1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Shared Memory Communications over RDMA (SMC-R) and RoCE 4 * 5 * AF_SMC protocol family socket handler keeping the AF_INET sock address type 6 * applies to SOCK_STREAM sockets only 7 * offers an alternative communication option for TCP-protocol sockets 8 * applicable with RoCE-cards only 9 * 10 * Initial restrictions: 11 * - support for alternate links postponed 12 * 13 * Copyright IBM Corp. 2016, 2018 14 * 15 * Author(s): Ursula Braun <ubraun@linux.vnet.ibm.com> 16 * based on prototype from Frank Blaschka 17 */ 18 19 #define pr_fmt(fmt) "smc: " fmt 20 21 #include <linux/module.h> 22 #include <linux/socket.h> 23 #include <linux/workqueue.h> 24 #include <linux/in.h> 25 #include <linux/sched/signal.h> 26 #include <linux/if_vlan.h> 27 #include <linux/rcupdate_wait.h> 28 #include <linux/ctype.h> 29 #include <linux/splice.h> 30 31 #include <net/sock.h> 32 #include <net/inet_common.h> 33 #if IS_ENABLED(CONFIG_IPV6) 34 #include <net/ipv6.h> 35 #endif 36 #include <net/tcp.h> 37 #include <net/smc.h> 38 #include <asm/ioctls.h> 39 40 #include <net/net_namespace.h> 41 #include <net/netns/generic.h> 42 #include "smc_netns.h" 43 44 #include "smc.h" 45 #include "smc_clc.h" 46 #include "smc_llc.h" 47 #include "smc_cdc.h" 48 #include "smc_core.h" 49 #include "smc_ib.h" 50 #include "smc_ism.h" 51 #include "smc_pnet.h" 52 #include "smc_netlink.h" 53 #include "smc_tx.h" 54 #include "smc_rx.h" 55 #include "smc_close.h" 56 #include "smc_stats.h" 57 #include "smc_tracepoint.h" 58 #include "smc_sysctl.h" 59 #include "smc_inet.h" 60 #include "smc_hs_bpf.h" 61 62 static DEFINE_MUTEX(smc_server_lgr_pending); /* serialize link group 63 * creation on server 64 */ 65 static DEFINE_MUTEX(smc_client_lgr_pending); /* serialize link group 66 * creation on client 67 */ 68 69 static struct workqueue_struct *smc_tcp_ls_wq; /* wq for tcp listen work */ 70 struct workqueue_struct *smc_hs_wq; /* wq for handshake work */ 71 struct workqueue_struct *smc_close_wq; /* wq for close work */ 72 73 static void smc_tcp_listen_work(struct work_struct *); 74 static void smc_connect_work(struct work_struct *); 75 76 int smc_nl_dump_hs_limitation(struct sk_buff *skb, struct netlink_callback *cb) 77 { 78 struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb); 79 void *hdr; 80 81 if (cb_ctx->pos[0]) 82 goto out; 83 84 hdr = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq, 85 &smc_gen_nl_family, NLM_F_MULTI, 86 SMC_NETLINK_DUMP_HS_LIMITATION); 87 if (!hdr) 88 return -ENOMEM; 89 90 if (nla_put_u8(skb, SMC_NLA_HS_LIMITATION_ENABLED, 91 sock_net(skb->sk)->smc.limit_smc_hs)) 92 goto err; 93 94 genlmsg_end(skb, hdr); 95 cb_ctx->pos[0] = 1; 96 out: 97 return skb->len; 98 err: 99 genlmsg_cancel(skb, hdr); 100 return -EMSGSIZE; 101 } 102 103 int smc_nl_enable_hs_limitation(struct sk_buff *skb, struct genl_info *info) 104 { 105 sock_net(skb->sk)->smc.limit_smc_hs = true; 106 return 0; 107 } 108 109 int smc_nl_disable_hs_limitation(struct sk_buff *skb, struct genl_info *info) 110 { 111 sock_net(skb->sk)->smc.limit_smc_hs = false; 112 return 0; 113 } 114 115 static void smc_set_keepalive(struct sock *sk, int val) 116 { 117 struct smc_sock *smc = smc_sk(sk); 118 119 smc->clcsock->sk->sk_prot->keepalive(smc->clcsock->sk, val); 120 } 121 122 static struct sock *smc_tcp_syn_recv_sock(const struct sock *sk, 123 struct sk_buff *skb, 124 struct request_sock *req, 125 struct dst_entry *dst, 126 struct request_sock *req_unhash, 127 bool *own_req, 128 void (*opt_child_init)(struct sock *newsk, 129 const struct sock *sk)) 130 { 131 struct smc_sock *smc; 132 struct sock *child; 133 134 rcu_read_lock(); 135 smc = smc_clcsock_user_data_rcu(sk); 136 if (!smc || !refcount_inc_not_zero(&smc->sk.sk_refcnt)) { 137 rcu_read_unlock(); 138 smc = NULL; 139 goto drop; 140 } 141 rcu_read_unlock(); 142 143 if (READ_ONCE(sk->sk_ack_backlog) + atomic_read(&smc->queued_smc_hs) > 144 sk->sk_max_ack_backlog) 145 goto drop; 146 147 if (sk_acceptq_is_full(&smc->sk)) { 148 NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); 149 goto drop; 150 } 151 152 /* passthrough to original syn recv sock fct */ 153 child = smc->ori_af_ops->syn_recv_sock(sk, skb, req, dst, req_unhash, 154 own_req, opt_child_init); 155 /* child must not inherit smc or its ops */ 156 if (child) { 157 rcu_assign_sk_user_data(child, NULL); 158 159 /* v4-mapped sockets don't inherit parent ops. Don't restore. */ 160 if (inet_csk(child)->icsk_af_ops == inet_csk(sk)->icsk_af_ops) 161 inet_csk(child)->icsk_af_ops = smc->ori_af_ops; 162 } 163 sock_put(&smc->sk); 164 return child; 165 166 drop: 167 dst_release(dst); 168 tcp_listendrop(sk); 169 if (smc) 170 sock_put(&smc->sk); 171 return NULL; 172 } 173 174 static bool smc_hs_congested(const struct sock *sk) 175 { 176 const struct smc_sock *smc; 177 178 smc = smc_clcsock_user_data(sk); 179 180 if (!smc) 181 return true; 182 183 if (workqueue_congested(WORK_CPU_UNBOUND, smc_hs_wq)) 184 return true; 185 186 return false; 187 } 188 189 struct smc_hashinfo smc_v4_hashinfo = { 190 .lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock), 191 }; 192 193 struct smc_hashinfo smc_v6_hashinfo = { 194 .lock = __RW_LOCK_UNLOCKED(smc_v6_hashinfo.lock), 195 }; 196 197 int smc_hash_sk(struct sock *sk) 198 { 199 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash; 200 struct hlist_head *head; 201 202 head = &h->ht; 203 204 write_lock_bh(&h->lock); 205 sk_add_node(sk, head); 206 write_unlock_bh(&h->lock); 207 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 208 209 return 0; 210 } 211 212 void smc_unhash_sk(struct sock *sk) 213 { 214 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash; 215 216 write_lock_bh(&h->lock); 217 if (sk_del_node_init(sk)) 218 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 219 write_unlock_bh(&h->lock); 220 } 221 222 /* This will be called before user really release sock_lock. So do the 223 * work which we didn't do because of user hold the sock_lock in the 224 * BH context 225 */ 226 void smc_release_cb(struct sock *sk) 227 { 228 struct smc_sock *smc = smc_sk(sk); 229 230 if (smc->conn.tx_in_release_sock) { 231 smc_tx_pending(&smc->conn); 232 smc->conn.tx_in_release_sock = false; 233 } 234 } 235 236 struct proto smc_proto = { 237 .name = "SMC", 238 .owner = THIS_MODULE, 239 .keepalive = smc_set_keepalive, 240 .hash = smc_hash_sk, 241 .unhash = smc_unhash_sk, 242 .release_cb = smc_release_cb, 243 .obj_size = sizeof(struct smc_sock), 244 .h.smc_hash = &smc_v4_hashinfo, 245 .slab_flags = SLAB_TYPESAFE_BY_RCU, 246 }; 247 EXPORT_SYMBOL_GPL(smc_proto); 248 249 struct proto smc_proto6 = { 250 .name = "SMC6", 251 .owner = THIS_MODULE, 252 .keepalive = smc_set_keepalive, 253 .hash = smc_hash_sk, 254 .unhash = smc_unhash_sk, 255 .release_cb = smc_release_cb, 256 .obj_size = sizeof(struct smc_sock), 257 .h.smc_hash = &smc_v6_hashinfo, 258 .slab_flags = SLAB_TYPESAFE_BY_RCU, 259 }; 260 EXPORT_SYMBOL_GPL(smc_proto6); 261 262 static void smc_fback_restore_callbacks(struct smc_sock *smc) 263 { 264 struct sock *clcsk = smc->clcsock->sk; 265 266 write_lock_bh(&clcsk->sk_callback_lock); 267 rcu_assign_sk_user_data(clcsk, NULL); 268 269 smc_clcsock_restore_cb(&clcsk->sk_state_change, &smc->clcsk_state_change); 270 smc_clcsock_restore_cb(&clcsk->sk_data_ready, &smc->clcsk_data_ready); 271 smc_clcsock_restore_cb(&clcsk->sk_write_space, &smc->clcsk_write_space); 272 smc_clcsock_restore_cb(&clcsk->sk_error_report, &smc->clcsk_error_report); 273 274 write_unlock_bh(&clcsk->sk_callback_lock); 275 } 276 277 static void smc_restore_fallback_changes(struct smc_sock *smc) 278 { 279 if (smc->clcsock->file) { /* non-accepted sockets have no file yet */ 280 smc->clcsock->file->private_data = smc->sk.sk_socket; 281 smc->clcsock->file = NULL; 282 smc_fback_restore_callbacks(smc); 283 } 284 } 285 286 static int __smc_release(struct smc_sock *smc) 287 { 288 struct sock *sk = &smc->sk; 289 int rc = 0; 290 291 if (!smc->use_fallback) { 292 rc = smc_close_active(smc); 293 smc_sock_set_flag(sk, SOCK_DEAD); 294 sk->sk_shutdown |= SHUTDOWN_MASK; 295 } else { 296 if (sk->sk_state != SMC_CLOSED) { 297 if (sk->sk_state != SMC_LISTEN && 298 sk->sk_state != SMC_INIT) 299 sock_put(sk); /* passive closing */ 300 if (sk->sk_state == SMC_LISTEN) { 301 /* wake up clcsock accept */ 302 rc = kernel_sock_shutdown(smc->clcsock, 303 SHUT_RDWR); 304 } 305 sk->sk_state = SMC_CLOSED; 306 sk->sk_state_change(sk); 307 } 308 smc_restore_fallback_changes(smc); 309 } 310 311 sk->sk_prot->unhash(sk); 312 313 if (sk->sk_state == SMC_CLOSED) { 314 if (smc->clcsock) { 315 release_sock(sk); 316 smc_clcsock_release(smc); 317 lock_sock(sk); 318 } 319 if (!smc->use_fallback) 320 smc_conn_free(&smc->conn); 321 } 322 323 return rc; 324 } 325 326 int smc_release(struct socket *sock) 327 { 328 struct sock *sk = sock->sk; 329 struct smc_sock *smc; 330 int old_state, rc = 0; 331 332 if (!sk) 333 goto out; 334 335 sock_hold(sk); /* sock_put below */ 336 smc = smc_sk(sk); 337 338 old_state = sk->sk_state; 339 340 /* cleanup for a dangling non-blocking connect */ 341 if (smc->connect_nonblock && old_state == SMC_INIT) 342 tcp_abort(smc->clcsock->sk, ECONNABORTED); 343 344 if (cancel_work_sync(&smc->connect_work)) 345 sock_put(&smc->sk); /* sock_hold in smc_connect for passive closing */ 346 347 if (sk->sk_state == SMC_LISTEN) 348 /* smc_close_non_accepted() is called and acquires 349 * sock lock for child sockets again 350 */ 351 lock_sock_nested(sk, SINGLE_DEPTH_NESTING); 352 else 353 lock_sock(sk); 354 355 if (old_state == SMC_INIT && sk->sk_state == SMC_ACTIVE && 356 !smc->use_fallback) 357 smc_close_active_abort(smc); 358 359 rc = __smc_release(smc); 360 361 /* detach socket */ 362 sock_orphan(sk); 363 sock->sk = NULL; 364 release_sock(sk); 365 366 sock_put(sk); /* sock_hold above */ 367 sock_put(sk); /* final sock_put */ 368 out: 369 return rc; 370 } 371 372 static void smc_destruct(struct sock *sk) 373 { 374 if (sk->sk_state != SMC_CLOSED) 375 return; 376 if (!sock_flag(sk, SOCK_DEAD)) 377 return; 378 switch (sk->sk_family) { 379 case AF_INET: 380 inet_sock_destruct(sk); 381 break; 382 #if IS_ENABLED(CONFIG_IPV6) 383 case AF_INET6: 384 inet6_sock_destruct(sk); 385 break; 386 #endif 387 } 388 } 389 390 static struct lock_class_key smc_key; 391 static struct lock_class_key smc_slock_key; 392 393 void smc_sk_init(struct net *net, struct sock *sk, int protocol) 394 { 395 struct smc_sock *smc = smc_sk(sk); 396 397 sk->sk_state = SMC_INIT; 398 sk->sk_destruct = smc_destruct; 399 sk->sk_protocol = protocol; 400 WRITE_ONCE(sk->sk_sndbuf, 2 * READ_ONCE(net->smc.sysctl_wmem)); 401 WRITE_ONCE(sk->sk_rcvbuf, 2 * READ_ONCE(net->smc.sysctl_rmem)); 402 INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work); 403 INIT_WORK(&smc->connect_work, smc_connect_work); 404 INIT_DELAYED_WORK(&smc->conn.tx_work, smc_tx_work); 405 INIT_LIST_HEAD(&smc->accept_q); 406 sock_lock_init_class_and_name(sk, "slock-AF_SMC", &smc_slock_key, 407 "sk_lock-AF_SMC", &smc_key); 408 spin_lock_init(&smc->accept_q_lock); 409 spin_lock_init(&smc->conn.send_lock); 410 sk->sk_prot->hash(sk); 411 mutex_init(&smc->clcsock_release_lock); 412 smc_init_saved_callbacks(smc); 413 smc->limit_smc_hs = net->smc.limit_smc_hs; 414 smc->use_fallback = false; /* assume rdma capability first */ 415 smc->fallback_rsn = 0; 416 smc_close_init(smc); 417 } 418 419 static struct sock *smc_sock_alloc(struct net *net, struct socket *sock, 420 int protocol) 421 { 422 struct proto *prot; 423 struct sock *sk; 424 425 prot = (protocol == SMCPROTO_SMC6) ? &smc_proto6 : &smc_proto; 426 sk = sk_alloc(net, PF_SMC, GFP_KERNEL, prot, 0); 427 if (!sk) 428 return NULL; 429 430 sock_init_data(sock, sk); /* sets sk_refcnt to 1 */ 431 smc_sk_init(net, sk, protocol); 432 433 return sk; 434 } 435 436 int smc_bind(struct socket *sock, struct sockaddr_unsized *uaddr, 437 int addr_len) 438 { 439 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 440 struct sock *sk = sock->sk; 441 struct smc_sock *smc; 442 int rc; 443 444 smc = smc_sk(sk); 445 446 /* replicate tests from inet_bind(), to be safe wrt. future changes */ 447 rc = -EINVAL; 448 if (addr_len < sizeof(struct sockaddr_in)) 449 goto out; 450 451 rc = -EAFNOSUPPORT; 452 if (addr->sin_family != AF_INET && 453 addr->sin_family != AF_INET6 && 454 addr->sin_family != AF_UNSPEC) 455 goto out; 456 /* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */ 457 if (addr->sin_family == AF_UNSPEC && 458 addr->sin_addr.s_addr != htonl(INADDR_ANY)) 459 goto out; 460 461 lock_sock(sk); 462 463 /* Check if socket is already active */ 464 rc = -EINVAL; 465 if (sk->sk_state != SMC_INIT || smc->connect_nonblock) 466 goto out_rel; 467 468 smc->clcsock->sk->sk_reuse = sk->sk_reuse; 469 smc->clcsock->sk->sk_reuseport = sk->sk_reuseport; 470 rc = kernel_bind(smc->clcsock, uaddr, addr_len); 471 472 out_rel: 473 release_sock(sk); 474 out: 475 return rc; 476 } 477 478 /* copy only relevant settings and flags of SOL_SOCKET level from smc to 479 * clc socket (since smc is not called for these options from net/core) 480 */ 481 482 #define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \ 483 (1UL << SOCK_KEEPOPEN) | \ 484 (1UL << SOCK_LINGER) | \ 485 (1UL << SOCK_BROADCAST) | \ 486 (1UL << SOCK_TIMESTAMP) | \ 487 (1UL << SOCK_DBG) | \ 488 (1UL << SOCK_RCVTSTAMP) | \ 489 (1UL << SOCK_RCVTSTAMPNS) | \ 490 (1UL << SOCK_LOCALROUTE) | \ 491 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \ 492 (1UL << SOCK_RXQ_OVFL) | \ 493 (1UL << SOCK_WIFI_STATUS) | \ 494 (1UL << SOCK_NOFCS) | \ 495 (1UL << SOCK_FILTER_LOCKED) | \ 496 (1UL << SOCK_TSTAMP_NEW)) 497 498 /* if set, use value set by setsockopt() - else use IPv4 or SMC sysctl value */ 499 static void smc_adjust_sock_bufsizes(struct sock *nsk, struct sock *osk, 500 unsigned long mask) 501 { 502 nsk->sk_userlocks = osk->sk_userlocks; 503 if (osk->sk_userlocks & SOCK_SNDBUF_LOCK) 504 nsk->sk_sndbuf = osk->sk_sndbuf; 505 if (osk->sk_userlocks & SOCK_RCVBUF_LOCK) 506 nsk->sk_rcvbuf = osk->sk_rcvbuf; 507 } 508 509 static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk, 510 unsigned long mask) 511 { 512 /* options we don't get control via setsockopt for */ 513 nsk->sk_type = osk->sk_type; 514 nsk->sk_sndtimeo = READ_ONCE(osk->sk_sndtimeo); 515 nsk->sk_rcvtimeo = READ_ONCE(osk->sk_rcvtimeo); 516 nsk->sk_mark = READ_ONCE(osk->sk_mark); 517 nsk->sk_priority = READ_ONCE(osk->sk_priority); 518 nsk->sk_rcvlowat = osk->sk_rcvlowat; 519 nsk->sk_bound_dev_if = osk->sk_bound_dev_if; 520 nsk->sk_err = osk->sk_err; 521 522 nsk->sk_flags &= ~mask; 523 nsk->sk_flags |= osk->sk_flags & mask; 524 525 smc_adjust_sock_bufsizes(nsk, osk, mask); 526 } 527 528 static void smc_copy_sock_settings_to_clc(struct smc_sock *smc) 529 { 530 smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC); 531 } 532 533 #define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \ 534 (1UL << SOCK_KEEPOPEN) | \ 535 (1UL << SOCK_LINGER) | \ 536 (1UL << SOCK_DBG)) 537 /* copy only settings and flags relevant for smc from clc to smc socket */ 538 static void smc_copy_sock_settings_to_smc(struct smc_sock *smc) 539 { 540 smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC); 541 } 542 543 /* register the new vzalloced sndbuf on all links */ 544 static int smcr_lgr_reg_sndbufs(struct smc_link *link, 545 struct smc_buf_desc *snd_desc) 546 { 547 struct smc_link_group *lgr = link->lgr; 548 int i, rc = 0; 549 550 if (!snd_desc->is_vm) 551 return -EINVAL; 552 553 /* protect against parallel smcr_link_reg_buf() */ 554 down_write(&lgr->llc_conf_mutex); 555 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 556 if (!smc_link_active(&lgr->lnk[i])) 557 continue; 558 rc = smcr_link_reg_buf(&lgr->lnk[i], snd_desc); 559 if (rc) 560 break; 561 } 562 up_write(&lgr->llc_conf_mutex); 563 return rc; 564 } 565 566 /* register the new rmb on all links */ 567 static int smcr_lgr_reg_rmbs(struct smc_link *link, 568 struct smc_buf_desc *rmb_desc) 569 { 570 struct smc_link_group *lgr = link->lgr; 571 bool do_slow = false; 572 int i, rc = 0; 573 574 rc = smc_llc_flow_initiate(lgr, SMC_LLC_FLOW_RKEY); 575 if (rc) 576 return rc; 577 578 down_read(&lgr->llc_conf_mutex); 579 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 580 if (!smc_link_active(&lgr->lnk[i])) 581 continue; 582 if (!rmb_desc->is_reg_mr[link->link_idx]) { 583 up_read(&lgr->llc_conf_mutex); 584 goto slow_path; 585 } 586 } 587 /* mr register already */ 588 goto fast_path; 589 slow_path: 590 do_slow = true; 591 /* protect against parallel smc_llc_cli_rkey_exchange() and 592 * parallel smcr_link_reg_buf() 593 */ 594 down_write(&lgr->llc_conf_mutex); 595 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 596 if (!smc_link_active(&lgr->lnk[i])) 597 continue; 598 rc = smcr_link_reg_buf(&lgr->lnk[i], rmb_desc); 599 if (rc) 600 goto out; 601 } 602 fast_path: 603 /* exchange confirm_rkey msg with peer */ 604 rc = smc_llc_do_confirm_rkey(link, rmb_desc); 605 if (rc) { 606 rc = -EFAULT; 607 goto out; 608 } 609 rmb_desc->is_conf_rkey = true; 610 out: 611 do_slow ? up_write(&lgr->llc_conf_mutex) : up_read(&lgr->llc_conf_mutex); 612 smc_llc_flow_stop(lgr, &lgr->llc_flow_lcl); 613 return rc; 614 } 615 616 static int smcr_clnt_conf_first_link(struct smc_sock *smc) 617 { 618 struct smc_link *link = smc->conn.lnk; 619 struct smc_llc_qentry *qentry; 620 int rc; 621 622 /* Receive CONFIRM LINK request from server over RoCE fabric. 623 * Increasing the client's timeout by twice as much as the server's 624 * timeout by default can temporarily avoid decline messages of 625 * both sides crossing or colliding 626 */ 627 qentry = smc_llc_wait(link->lgr, NULL, 2 * SMC_LLC_WAIT_TIME, 628 SMC_LLC_CONFIRM_LINK); 629 if (!qentry) { 630 struct smc_clc_msg_decline dclc; 631 632 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 633 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 634 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 635 } 636 smc_llc_save_peer_uid(qentry); 637 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_REQ); 638 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl); 639 if (rc) 640 return SMC_CLC_DECL_RMBE_EC; 641 642 rc = smc_ib_modify_qp_rts(link); 643 if (rc) 644 return SMC_CLC_DECL_ERR_RDYLNK; 645 646 smc_wr_remember_qp_attr(link); 647 648 /* reg the sndbuf if it was vzalloced */ 649 if (smc->conn.sndbuf_desc->is_vm) { 650 if (smcr_link_reg_buf(link, smc->conn.sndbuf_desc)) 651 return SMC_CLC_DECL_ERR_REGBUF; 652 } 653 654 /* reg the rmb */ 655 if (smcr_link_reg_buf(link, smc->conn.rmb_desc)) 656 return SMC_CLC_DECL_ERR_REGBUF; 657 658 /* confirm_rkey is implicit on 1st contact */ 659 smc->conn.rmb_desc->is_conf_rkey = true; 660 661 /* send CONFIRM LINK response over RoCE fabric */ 662 rc = smc_llc_send_confirm_link(link, SMC_LLC_RESP); 663 if (rc < 0) 664 return SMC_CLC_DECL_TIMEOUT_CL; 665 666 smc_llc_link_active(link); 667 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE); 668 669 if (link->lgr->max_links > 1) { 670 /* optional 2nd link, receive ADD LINK request from server */ 671 qentry = smc_llc_wait(link->lgr, NULL, SMC_LLC_WAIT_TIME, 672 SMC_LLC_ADD_LINK); 673 if (!qentry) { 674 struct smc_clc_msg_decline dclc; 675 676 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 677 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 678 if (rc == -EAGAIN) 679 rc = 0; /* no DECLINE received, go with one link */ 680 return rc; 681 } 682 smc_llc_flow_qentry_clr(&link->lgr->llc_flow_lcl); 683 smc_llc_cli_add_link(link, qentry); 684 } 685 return 0; 686 } 687 688 static bool smc_isascii(char *hostname) 689 { 690 int i; 691 692 for (i = 0; i < SMC_MAX_HOSTNAME_LEN; i++) 693 if (!isascii(hostname[i])) 694 return false; 695 return true; 696 } 697 698 static void smc_conn_save_peer_info_fce(struct smc_sock *smc, 699 struct smc_clc_msg_accept_confirm *clc) 700 { 701 struct smc_clc_first_contact_ext *fce; 702 int clc_v2_len; 703 704 if (clc->hdr.version == SMC_V1 || 705 !(clc->hdr.typev2 & SMC_FIRST_CONTACT_MASK)) 706 return; 707 708 if (smc->conn.lgr->is_smcd) { 709 memcpy(smc->conn.lgr->negotiated_eid, clc->d1.eid, 710 SMC_MAX_EID_LEN); 711 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm, d1); 712 } else { 713 memcpy(smc->conn.lgr->negotiated_eid, clc->r1.eid, 714 SMC_MAX_EID_LEN); 715 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm, r1); 716 } 717 fce = (struct smc_clc_first_contact_ext *)(((u8 *)clc) + clc_v2_len); 718 smc->conn.lgr->peer_os = fce->os_type; 719 smc->conn.lgr->peer_smc_release = fce->release; 720 if (smc_isascii(fce->hostname)) 721 memcpy(smc->conn.lgr->peer_hostname, fce->hostname, 722 SMC_MAX_HOSTNAME_LEN); 723 } 724 725 static void smcr_conn_save_peer_info(struct smc_sock *smc, 726 struct smc_clc_msg_accept_confirm *clc) 727 { 728 int bufsize = smc_uncompress_bufsize(clc->r0.rmbe_size); 729 730 smc->conn.peer_rmbe_idx = clc->r0.rmbe_idx; 731 smc->conn.local_tx_ctrl.token = ntohl(clc->r0.rmbe_alert_token); 732 smc->conn.peer_rmbe_size = bufsize; 733 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 734 smc->conn.tx_off = bufsize * (smc->conn.peer_rmbe_idx - 1); 735 } 736 737 static void smcd_conn_save_peer_info(struct smc_sock *smc, 738 struct smc_clc_msg_accept_confirm *clc) 739 { 740 int bufsize = smc_uncompress_bufsize(clc->d0.dmbe_size); 741 742 smc->conn.peer_rmbe_idx = clc->d0.dmbe_idx; 743 smc->conn.peer_token = ntohll(clc->d0.token); 744 /* msg header takes up space in the buffer */ 745 smc->conn.peer_rmbe_size = bufsize - sizeof(struct smcd_cdc_msg); 746 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 747 smc->conn.tx_off = bufsize * smc->conn.peer_rmbe_idx; 748 } 749 750 static void smc_conn_save_peer_info(struct smc_sock *smc, 751 struct smc_clc_msg_accept_confirm *clc) 752 { 753 if (smc->conn.lgr->is_smcd) 754 smcd_conn_save_peer_info(smc, clc); 755 else 756 smcr_conn_save_peer_info(smc, clc); 757 smc_conn_save_peer_info_fce(smc, clc); 758 } 759 760 static void smc_link_save_peer_info(struct smc_link *link, 761 struct smc_clc_msg_accept_confirm *clc, 762 struct smc_init_info *ini) 763 { 764 link->peer_qpn = ntoh24(clc->r0.qpn); 765 memcpy(link->peer_gid, ini->peer_gid, SMC_GID_SIZE); 766 memcpy(link->peer_mac, ini->peer_mac, sizeof(link->peer_mac)); 767 link->peer_psn = ntoh24(clc->r0.psn); 768 link->peer_mtu = clc->r0.qp_mtu; 769 } 770 771 static void smc_stat_inc_fback_rsn_cnt(struct smc_sock *smc, 772 struct smc_stats_fback *fback_arr) 773 { 774 int cnt; 775 776 for (cnt = 0; cnt < SMC_MAX_FBACK_RSN_CNT; cnt++) { 777 if (fback_arr[cnt].fback_code == smc->fallback_rsn) { 778 fback_arr[cnt].count++; 779 break; 780 } 781 if (!fback_arr[cnt].fback_code) { 782 fback_arr[cnt].fback_code = smc->fallback_rsn; 783 fback_arr[cnt].count++; 784 break; 785 } 786 } 787 } 788 789 static void smc_stat_fallback(struct smc_sock *smc) 790 { 791 struct net *net = sock_net(&smc->sk); 792 793 mutex_lock(&net->smc.mutex_fback_rsn); 794 if (smc->listen_smc) { 795 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->srv); 796 net->smc.fback_rsn->srv_fback_cnt++; 797 } else { 798 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->clnt); 799 net->smc.fback_rsn->clnt_fback_cnt++; 800 } 801 mutex_unlock(&net->smc.mutex_fback_rsn); 802 } 803 804 /* must be called under rcu read lock */ 805 static void smc_fback_wakeup_waitqueue(struct smc_sock *smc, void *key) 806 { 807 struct socket_wq *wq; 808 __poll_t flags; 809 810 wq = rcu_dereference(smc->sk.sk_wq); 811 if (!skwq_has_sleeper(wq)) 812 return; 813 814 /* wake up smc sk->sk_wq */ 815 if (!key) { 816 /* sk_state_change */ 817 wake_up_interruptible_all(&wq->wait); 818 } else { 819 flags = key_to_poll(key); 820 if (flags & (EPOLLIN | EPOLLOUT)) 821 /* sk_data_ready or sk_write_space */ 822 wake_up_interruptible_sync_poll(&wq->wait, flags); 823 else if (flags & EPOLLERR) 824 /* sk_error_report */ 825 wake_up_interruptible_poll(&wq->wait, flags); 826 } 827 } 828 829 static int smc_fback_mark_woken(wait_queue_entry_t *wait, 830 unsigned int mode, int sync, void *key) 831 { 832 struct smc_mark_woken *mark = 833 container_of(wait, struct smc_mark_woken, wait_entry); 834 835 mark->woken = true; 836 mark->key = key; 837 return 0; 838 } 839 840 static void smc_fback_forward_wakeup(struct smc_sock *smc, struct sock *clcsk, 841 void (*clcsock_callback)(struct sock *sk)) 842 { 843 struct smc_mark_woken mark = { .woken = false }; 844 struct socket_wq *wq; 845 846 init_waitqueue_func_entry(&mark.wait_entry, 847 smc_fback_mark_woken); 848 rcu_read_lock(); 849 wq = rcu_dereference(clcsk->sk_wq); 850 if (!wq) 851 goto out; 852 add_wait_queue(sk_sleep(clcsk), &mark.wait_entry); 853 clcsock_callback(clcsk); 854 remove_wait_queue(sk_sleep(clcsk), &mark.wait_entry); 855 856 if (mark.woken) 857 smc_fback_wakeup_waitqueue(smc, mark.key); 858 out: 859 rcu_read_unlock(); 860 } 861 862 static void smc_fback_state_change(struct sock *clcsk) 863 { 864 struct smc_sock *smc; 865 866 read_lock_bh(&clcsk->sk_callback_lock); 867 smc = smc_clcsock_user_data(clcsk); 868 if (smc) 869 smc_fback_forward_wakeup(smc, clcsk, 870 smc->clcsk_state_change); 871 read_unlock_bh(&clcsk->sk_callback_lock); 872 } 873 874 static void smc_fback_data_ready(struct sock *clcsk) 875 { 876 struct smc_sock *smc; 877 878 read_lock_bh(&clcsk->sk_callback_lock); 879 smc = smc_clcsock_user_data(clcsk); 880 if (smc) 881 smc_fback_forward_wakeup(smc, clcsk, 882 smc->clcsk_data_ready); 883 read_unlock_bh(&clcsk->sk_callback_lock); 884 } 885 886 static void smc_fback_write_space(struct sock *clcsk) 887 { 888 struct smc_sock *smc; 889 890 read_lock_bh(&clcsk->sk_callback_lock); 891 smc = smc_clcsock_user_data(clcsk); 892 if (smc) 893 smc_fback_forward_wakeup(smc, clcsk, 894 smc->clcsk_write_space); 895 read_unlock_bh(&clcsk->sk_callback_lock); 896 } 897 898 static void smc_fback_error_report(struct sock *clcsk) 899 { 900 struct smc_sock *smc; 901 902 read_lock_bh(&clcsk->sk_callback_lock); 903 smc = smc_clcsock_user_data(clcsk); 904 if (smc) 905 smc_fback_forward_wakeup(smc, clcsk, 906 smc->clcsk_error_report); 907 read_unlock_bh(&clcsk->sk_callback_lock); 908 } 909 910 static void smc_fback_replace_callbacks(struct smc_sock *smc) 911 { 912 struct sock *clcsk = smc->clcsock->sk; 913 914 write_lock_bh(&clcsk->sk_callback_lock); 915 __rcu_assign_sk_user_data_with_flags(clcsk, smc, SK_USER_DATA_NOCOPY); 916 917 smc_clcsock_replace_cb(&clcsk->sk_state_change, smc_fback_state_change, 918 &smc->clcsk_state_change); 919 smc_clcsock_replace_cb(&clcsk->sk_data_ready, smc_fback_data_ready, 920 &smc->clcsk_data_ready); 921 smc_clcsock_replace_cb(&clcsk->sk_write_space, smc_fback_write_space, 922 &smc->clcsk_write_space); 923 smc_clcsock_replace_cb(&clcsk->sk_error_report, smc_fback_error_report, 924 &smc->clcsk_error_report); 925 926 write_unlock_bh(&clcsk->sk_callback_lock); 927 } 928 929 static int smc_switch_to_fallback(struct smc_sock *smc, int reason_code) 930 { 931 int rc = 0; 932 933 mutex_lock(&smc->clcsock_release_lock); 934 if (!smc->clcsock) { 935 rc = -EBADF; 936 goto out; 937 } 938 939 smc->use_fallback = true; 940 smc->fallback_rsn = reason_code; 941 smc_stat_fallback(smc); 942 trace_smc_switch_to_fallback(smc, reason_code); 943 if (smc->sk.sk_socket && smc->sk.sk_socket->file) { 944 smc->clcsock->file = smc->sk.sk_socket->file; 945 smc->clcsock->file->private_data = smc->clcsock; 946 smc->clcsock->wq.fasync_list = 947 smc->sk.sk_socket->wq.fasync_list; 948 smc->sk.sk_socket->wq.fasync_list = NULL; 949 950 /* There might be some wait entries remaining 951 * in smc sk->sk_wq and they should be woken up 952 * as clcsock's wait queue is woken up. 953 */ 954 smc_fback_replace_callbacks(smc); 955 } 956 out: 957 mutex_unlock(&smc->clcsock_release_lock); 958 return rc; 959 } 960 961 /* fall back during connect */ 962 static int smc_connect_fallback(struct smc_sock *smc, int reason_code) 963 { 964 struct net *net = sock_net(&smc->sk); 965 int rc = 0; 966 967 rc = smc_switch_to_fallback(smc, reason_code); 968 if (rc) { /* fallback fails */ 969 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 970 if (smc->sk.sk_state == SMC_INIT) 971 sock_put(&smc->sk); /* passive closing */ 972 return rc; 973 } 974 smc_copy_sock_settings_to_clc(smc); 975 smc->connect_nonblock = 0; 976 if (smc->sk.sk_state == SMC_INIT) 977 smc->sk.sk_state = SMC_ACTIVE; 978 return 0; 979 } 980 981 /* decline and fall back during connect */ 982 static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code, 983 u8 version) 984 { 985 struct net *net = sock_net(&smc->sk); 986 int rc; 987 988 if (reason_code < 0) { /* error, fallback is not possible */ 989 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 990 if (smc->sk.sk_state == SMC_INIT) 991 sock_put(&smc->sk); /* passive closing */ 992 return reason_code; 993 } 994 if (reason_code != SMC_CLC_DECL_PEERDECL) { 995 rc = smc_clc_send_decline(smc, reason_code, version); 996 if (rc < 0) { 997 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 998 if (smc->sk.sk_state == SMC_INIT) 999 sock_put(&smc->sk); /* passive closing */ 1000 return rc; 1001 } 1002 } 1003 return smc_connect_fallback(smc, reason_code); 1004 } 1005 1006 static void smc_conn_abort(struct smc_sock *smc, int local_first) 1007 { 1008 struct smc_connection *conn = &smc->conn; 1009 struct smc_link_group *lgr = conn->lgr; 1010 bool lgr_valid = false; 1011 1012 if (smc_conn_lgr_valid(conn)) 1013 lgr_valid = true; 1014 1015 smc_conn_free(conn); 1016 if (local_first && lgr_valid) 1017 smc_lgr_cleanup_early(lgr); 1018 } 1019 1020 /* check if there is a rdma device available for this connection. */ 1021 /* called for connect and listen */ 1022 static int smc_find_rdma_device(struct smc_sock *smc, struct smc_init_info *ini) 1023 { 1024 /* PNET table look up: search active ib_device and port 1025 * within same PNETID that also contains the ethernet device 1026 * used for the internal TCP socket 1027 */ 1028 smc_pnet_find_roce_resource(smc->clcsock->sk, ini); 1029 if (!ini->check_smcrv2 && !ini->ib_dev) 1030 return SMC_CLC_DECL_NOSMCRDEV; 1031 if (ini->check_smcrv2 && !ini->smcrv2.ib_dev_v2) 1032 return SMC_CLC_DECL_NOSMCRDEV; 1033 return 0; 1034 } 1035 1036 /* check if there is an ISM device available for this connection. */ 1037 /* called for connect and listen */ 1038 static int smc_find_ism_device(struct smc_sock *smc, struct smc_init_info *ini) 1039 { 1040 /* Find ISM device with same PNETID as connecting interface */ 1041 smc_pnet_find_ism_resource(smc->clcsock->sk, ini); 1042 if (!ini->ism_dev[0]) 1043 return SMC_CLC_DECL_NOSMCDDEV; 1044 else 1045 ini->ism_chid[0] = smc_ism_get_chid(ini->ism_dev[0]); 1046 return 0; 1047 } 1048 1049 /* is chid unique for the ism devices that are already determined? */ 1050 static bool smc_find_ism_v2_is_unique_chid(u16 chid, struct smc_init_info *ini, 1051 int cnt) 1052 { 1053 int i = (!ini->ism_dev[0]) ? 1 : 0; 1054 1055 for (; i < cnt; i++) 1056 if (ini->ism_chid[i] == chid) 1057 return false; 1058 return true; 1059 } 1060 1061 /* determine possible V2 ISM devices (either without PNETID or with PNETID plus 1062 * PNETID matching net_device) 1063 */ 1064 static int smc_find_ism_v2_device_clnt(struct smc_sock *smc, 1065 struct smc_init_info *ini) 1066 { 1067 int rc = SMC_CLC_DECL_NOSMCDDEV; 1068 struct smcd_dev *smcd; 1069 int i = 1, entry = 1; 1070 bool is_emulated; 1071 u16 chid; 1072 1073 if (smcd_indicated(ini->smc_type_v1)) 1074 rc = 0; /* already initialized for V1 */ 1075 mutex_lock(&smcd_dev_list.mutex); 1076 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 1077 if (smcd->going_away || smcd == ini->ism_dev[0]) 1078 continue; 1079 chid = smc_ism_get_chid(smcd); 1080 if (!smc_find_ism_v2_is_unique_chid(chid, ini, i)) 1081 continue; 1082 is_emulated = __smc_ism_is_emulated(chid); 1083 if (!smc_pnet_is_pnetid_set(smcd->pnetid) || 1084 smc_pnet_is_ndev_pnetid(sock_net(&smc->sk), smcd->pnetid)) { 1085 if (is_emulated && entry == SMCD_CLC_MAX_V2_GID_ENTRIES) 1086 /* It's the last GID-CHID entry left in CLC 1087 * Proposal SMC-Dv2 extension, but an Emulated- 1088 * ISM device will take two entries. So give 1089 * up it and try the next potential ISM device. 1090 */ 1091 continue; 1092 ini->ism_dev[i] = smcd; 1093 ini->ism_chid[i] = chid; 1094 ini->is_smcd = true; 1095 rc = 0; 1096 i++; 1097 entry = is_emulated ? entry + 2 : entry + 1; 1098 if (entry > SMCD_CLC_MAX_V2_GID_ENTRIES) 1099 break; 1100 } 1101 } 1102 mutex_unlock(&smcd_dev_list.mutex); 1103 ini->ism_offered_cnt = i - 1; 1104 if (!ini->ism_dev[0] && !ini->ism_dev[1]) 1105 ini->smcd_version = 0; 1106 1107 return rc; 1108 } 1109 1110 /* Check for VLAN ID and register it on ISM device just for CLC handshake */ 1111 static int smc_connect_ism_vlan_setup(struct smc_init_info *ini) 1112 { 1113 if (ini->vlan_id && smc_ism_get_vlan(ini->ism_dev[0], ini->vlan_id)) 1114 return SMC_CLC_DECL_ISMVLANERR; 1115 return 0; 1116 } 1117 1118 static int smc_find_proposal_devices(struct smc_sock *smc, 1119 struct smc_init_info *ini) 1120 { 1121 int rc = 0; 1122 1123 /* check if there is an ism device available */ 1124 if (!(ini->smcd_version & SMC_V1) || 1125 smc_find_ism_device(smc, ini) || 1126 smc_connect_ism_vlan_setup(ini)) 1127 ini->smcd_version &= ~SMC_V1; 1128 /* else ISM V1 is supported for this connection */ 1129 1130 /* check if there is an rdma device available */ 1131 if (!(ini->smcr_version & SMC_V1) || 1132 smc_find_rdma_device(smc, ini)) 1133 ini->smcr_version &= ~SMC_V1; 1134 /* else RDMA is supported for this connection */ 1135 1136 ini->smc_type_v1 = smc_indicated_type(ini->smcd_version & SMC_V1, 1137 ini->smcr_version & SMC_V1); 1138 1139 /* check if there is an ism v2 device available */ 1140 if (!(ini->smcd_version & SMC_V2) || 1141 !smc_ism_is_v2_capable() || 1142 smc_find_ism_v2_device_clnt(smc, ini)) 1143 ini->smcd_version &= ~SMC_V2; 1144 1145 /* check if there is an rdma v2 device available */ 1146 ini->check_smcrv2 = true; 1147 ini->smcrv2.saddr = smc->clcsock->sk->sk_rcv_saddr; 1148 if (!(ini->smcr_version & SMC_V2) || 1149 #if IS_ENABLED(CONFIG_IPV6) 1150 (smc->clcsock->sk->sk_family == AF_INET6 && 1151 !ipv6_addr_v4mapped(&smc->clcsock->sk->sk_v6_rcv_saddr)) || 1152 #endif 1153 !smc_clc_ueid_count() || 1154 smc_find_rdma_device(smc, ini)) 1155 ini->smcr_version &= ~SMC_V2; 1156 ini->check_smcrv2 = false; 1157 1158 ini->smc_type_v2 = smc_indicated_type(ini->smcd_version & SMC_V2, 1159 ini->smcr_version & SMC_V2); 1160 1161 /* if neither ISM nor RDMA are supported, fallback */ 1162 if (ini->smc_type_v1 == SMC_TYPE_N && ini->smc_type_v2 == SMC_TYPE_N) 1163 rc = SMC_CLC_DECL_NOSMCDEV; 1164 1165 return rc; 1166 } 1167 1168 /* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is 1169 * used, the VLAN ID will be registered again during the connection setup. 1170 */ 1171 static int smc_connect_ism_vlan_cleanup(struct smc_init_info *ini) 1172 { 1173 if (!smcd_indicated(ini->smc_type_v1)) 1174 return 0; 1175 if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev[0], ini->vlan_id)) 1176 return SMC_CLC_DECL_CNFERR; 1177 return 0; 1178 } 1179 1180 #define SMC_CLC_MAX_ACCEPT_LEN \ 1181 (sizeof(struct smc_clc_msg_accept_confirm) + \ 1182 sizeof(struct smc_clc_first_contact_ext_v2x) + \ 1183 sizeof(struct smc_clc_msg_trail)) 1184 1185 /* CLC handshake during connect */ 1186 static int smc_connect_clc(struct smc_sock *smc, 1187 struct smc_clc_msg_accept_confirm *aclc, 1188 struct smc_init_info *ini) 1189 { 1190 int rc = 0; 1191 1192 /* do inband token exchange */ 1193 rc = smc_clc_send_proposal(smc, ini); 1194 if (rc) 1195 return rc; 1196 /* receive SMC Accept CLC message */ 1197 return smc_clc_wait_msg(smc, aclc, SMC_CLC_MAX_ACCEPT_LEN, 1198 SMC_CLC_ACCEPT, CLC_WAIT_TIME); 1199 } 1200 1201 void smc_fill_gid_list(struct smc_link_group *lgr, 1202 struct smc_gidlist *gidlist, 1203 struct smc_ib_device *known_dev, u8 *known_gid) 1204 { 1205 struct smc_init_info *alt_ini = NULL; 1206 1207 memset(gidlist, 0, sizeof(*gidlist)); 1208 memcpy(gidlist->list[gidlist->len++], known_gid, SMC_GID_SIZE); 1209 1210 alt_ini = kzalloc_obj(*alt_ini); 1211 if (!alt_ini) 1212 goto out; 1213 1214 alt_ini->vlan_id = lgr->vlan_id; 1215 alt_ini->check_smcrv2 = true; 1216 alt_ini->smcrv2.saddr = lgr->saddr; 1217 smc_pnet_find_alt_roce(lgr, alt_ini, known_dev); 1218 1219 if (!alt_ini->smcrv2.ib_dev_v2) 1220 goto out; 1221 1222 memcpy(gidlist->list[gidlist->len++], alt_ini->smcrv2.ib_gid_v2, 1223 SMC_GID_SIZE); 1224 1225 out: 1226 kfree(alt_ini); 1227 } 1228 1229 static int smc_connect_rdma_v2_prepare(struct smc_sock *smc, 1230 struct smc_clc_msg_accept_confirm *aclc, 1231 struct smc_init_info *ini) 1232 { 1233 struct smc_clc_first_contact_ext *fce = 1234 smc_get_clc_first_contact_ext(aclc, false); 1235 struct net *net = sock_net(&smc->sk); 1236 int rc; 1237 1238 if (!ini->first_contact_peer || aclc->hdr.version == SMC_V1) 1239 return 0; 1240 1241 if (fce->v2_direct) { 1242 memcpy(ini->smcrv2.nexthop_mac, &aclc->r0.lcl.mac, ETH_ALEN); 1243 ini->smcrv2.uses_gateway = false; 1244 } else { 1245 if (smc_ib_find_route(net, smc->clcsock->sk->sk_rcv_saddr, 1246 smc_ib_gid_to_ipv4(aclc->r0.lcl.gid), 1247 ini->smcrv2.nexthop_mac, 1248 &ini->smcrv2.uses_gateway)) 1249 return SMC_CLC_DECL_NOROUTE; 1250 if (!ini->smcrv2.uses_gateway) { 1251 /* mismatch: peer claims indirect, but its direct */ 1252 return SMC_CLC_DECL_NOINDIRECT; 1253 } 1254 } 1255 1256 ini->release_nr = fce->release; 1257 rc = smc_clc_clnt_v2x_features_validate(fce, ini); 1258 if (rc) 1259 return rc; 1260 1261 return 0; 1262 } 1263 1264 /* setup for RDMA connection of client */ 1265 static int smc_connect_rdma(struct smc_sock *smc, 1266 struct smc_clc_msg_accept_confirm *aclc, 1267 struct smc_init_info *ini) 1268 { 1269 int i, reason_code = 0; 1270 struct smc_link *link; 1271 u8 *eid = NULL; 1272 1273 ini->is_smcd = false; 1274 ini->ib_clcqpn = ntoh24(aclc->r0.qpn); 1275 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK; 1276 memcpy(ini->peer_systemid, aclc->r0.lcl.id_for_peer, SMC_SYSTEMID_LEN); 1277 memcpy(ini->peer_gid, aclc->r0.lcl.gid, SMC_GID_SIZE); 1278 memcpy(ini->peer_mac, aclc->r0.lcl.mac, ETH_ALEN); 1279 ini->max_conns = SMC_CONN_PER_LGR_MAX; 1280 ini->max_links = SMC_LINKS_ADD_LNK_MAX; 1281 1282 reason_code = smc_connect_rdma_v2_prepare(smc, aclc, ini); 1283 if (reason_code) 1284 return reason_code; 1285 1286 mutex_lock(&smc_client_lgr_pending); 1287 reason_code = smc_conn_create(smc, ini); 1288 if (reason_code) { 1289 mutex_unlock(&smc_client_lgr_pending); 1290 return reason_code; 1291 } 1292 1293 smc_conn_save_peer_info(smc, aclc); 1294 1295 if (ini->first_contact_local) { 1296 link = smc->conn.lnk; 1297 } else { 1298 /* set link that was assigned by server */ 1299 link = NULL; 1300 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 1301 struct smc_link *l = &smc->conn.lgr->lnk[i]; 1302 1303 if (l->peer_qpn == ntoh24(aclc->r0.qpn) && 1304 !memcmp(l->peer_gid, &aclc->r0.lcl.gid, 1305 SMC_GID_SIZE) && 1306 (aclc->hdr.version > SMC_V1 || 1307 !memcmp(l->peer_mac, &aclc->r0.lcl.mac, 1308 sizeof(l->peer_mac)))) { 1309 link = l; 1310 break; 1311 } 1312 } 1313 if (!link) { 1314 reason_code = SMC_CLC_DECL_NOSRVLINK; 1315 goto connect_abort; 1316 } 1317 smc_switch_link_and_count(&smc->conn, link); 1318 } 1319 1320 /* create send buffer and rmb */ 1321 if (smc_buf_create(smc, false)) { 1322 reason_code = SMC_CLC_DECL_MEM; 1323 goto connect_abort; 1324 } 1325 1326 if (ini->first_contact_local) 1327 smc_link_save_peer_info(link, aclc, ini); 1328 1329 if (smc_rmb_rtoken_handling(&smc->conn, link, aclc)) { 1330 reason_code = SMC_CLC_DECL_ERR_RTOK; 1331 goto connect_abort; 1332 } 1333 1334 smc_rx_init(smc); 1335 1336 if (ini->first_contact_local) { 1337 if (smc_ib_ready_link(link)) { 1338 reason_code = SMC_CLC_DECL_ERR_RDYLNK; 1339 goto connect_abort; 1340 } 1341 } else { 1342 /* reg sendbufs if they were vzalloced */ 1343 if (smc->conn.sndbuf_desc->is_vm) { 1344 if (smcr_lgr_reg_sndbufs(link, smc->conn.sndbuf_desc)) { 1345 reason_code = SMC_CLC_DECL_ERR_REGBUF; 1346 goto connect_abort; 1347 } 1348 } 1349 if (smcr_lgr_reg_rmbs(link, smc->conn.rmb_desc)) { 1350 reason_code = SMC_CLC_DECL_ERR_REGBUF; 1351 goto connect_abort; 1352 } 1353 } 1354 1355 if (aclc->hdr.version > SMC_V1) { 1356 eid = aclc->r1.eid; 1357 if (ini->first_contact_local) 1358 smc_fill_gid_list(link->lgr, &ini->smcrv2.gidlist, 1359 link->smcibdev, link->gid); 1360 } 1361 1362 reason_code = smc_clc_send_confirm(smc, ini->first_contact_local, 1363 aclc->hdr.version, eid, ini); 1364 if (reason_code) 1365 goto connect_abort; 1366 1367 smc_tx_init(smc); 1368 1369 if (ini->first_contact_local) { 1370 /* QP confirmation over RoCE fabric */ 1371 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 1372 reason_code = smcr_clnt_conf_first_link(smc); 1373 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 1374 if (reason_code) 1375 goto connect_abort; 1376 } 1377 mutex_unlock(&smc_client_lgr_pending); 1378 1379 smc_copy_sock_settings_to_clc(smc); 1380 smc->connect_nonblock = 0; 1381 if (smc->sk.sk_state == SMC_INIT) 1382 smc->sk.sk_state = SMC_ACTIVE; 1383 1384 return 0; 1385 connect_abort: 1386 smc_conn_abort(smc, ini->first_contact_local); 1387 mutex_unlock(&smc_client_lgr_pending); 1388 smc->connect_nonblock = 0; 1389 1390 return reason_code; 1391 } 1392 1393 /* The server has chosen one of the proposed ISM devices for the communication. 1394 * Determine from the CHID of the received CLC ACCEPT the ISM device chosen. 1395 */ 1396 static int 1397 smc_v2_determine_accepted_chid(struct smc_clc_msg_accept_confirm *aclc, 1398 struct smc_init_info *ini) 1399 { 1400 int i; 1401 1402 for (i = 0; i < ini->ism_offered_cnt + 1; i++) { 1403 if (ini->ism_dev[i] && 1404 ini->ism_chid[i] == ntohs(aclc->d1.chid)) { 1405 ini->ism_selected = i; 1406 return 0; 1407 } 1408 } 1409 1410 return -EPROTO; 1411 } 1412 1413 /* setup for ISM connection of client */ 1414 static int smc_connect_ism(struct smc_sock *smc, 1415 struct smc_clc_msg_accept_confirm *aclc, 1416 struct smc_init_info *ini) 1417 { 1418 u8 *eid = NULL; 1419 int rc = 0; 1420 1421 ini->is_smcd = true; 1422 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK; 1423 1424 if (aclc->hdr.version == SMC_V2) { 1425 if (ini->first_contact_peer) { 1426 struct smc_clc_first_contact_ext *fce = 1427 smc_get_clc_first_contact_ext(aclc, true); 1428 1429 ini->release_nr = fce->release; 1430 rc = smc_clc_clnt_v2x_features_validate(fce, ini); 1431 if (rc) 1432 return rc; 1433 } 1434 1435 rc = smc_v2_determine_accepted_chid(aclc, ini); 1436 if (rc) 1437 return rc; 1438 1439 if (__smc_ism_is_emulated(ini->ism_chid[ini->ism_selected])) 1440 ini->ism_peer_gid[ini->ism_selected].gid_ext = 1441 ntohll(aclc->d1.gid_ext); 1442 /* for non-Emulated-ISM devices, peer gid_ext remains 0. */ 1443 } 1444 ini->ism_peer_gid[ini->ism_selected].gid = ntohll(aclc->d0.gid); 1445 1446 /* there is only one lgr role for SMC-D; use server lock */ 1447 mutex_lock(&smc_server_lgr_pending); 1448 rc = smc_conn_create(smc, ini); 1449 if (rc) { 1450 mutex_unlock(&smc_server_lgr_pending); 1451 return rc; 1452 } 1453 1454 /* Create send and receive buffers */ 1455 rc = smc_buf_create(smc, true); 1456 if (rc) { 1457 rc = (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : SMC_CLC_DECL_MEM; 1458 goto connect_abort; 1459 } 1460 1461 smc_conn_save_peer_info(smc, aclc); 1462 1463 if (smc_ism_support_dmb_nocopy(smc->conn.lgr->smcd)) { 1464 rc = smcd_buf_attach(smc); 1465 if (rc) { 1466 rc = SMC_CLC_DECL_MEM; /* try to fallback */ 1467 goto connect_abort; 1468 } 1469 } 1470 smc_rx_init(smc); 1471 smc_tx_init(smc); 1472 1473 if (aclc->hdr.version > SMC_V1) 1474 eid = aclc->d1.eid; 1475 1476 rc = smc_clc_send_confirm(smc, ini->first_contact_local, 1477 aclc->hdr.version, eid, ini); 1478 if (rc) 1479 goto connect_abort; 1480 mutex_unlock(&smc_server_lgr_pending); 1481 1482 smc_copy_sock_settings_to_clc(smc); 1483 smc->connect_nonblock = 0; 1484 if (smc->sk.sk_state == SMC_INIT) 1485 smc->sk.sk_state = SMC_ACTIVE; 1486 1487 return 0; 1488 connect_abort: 1489 smc_conn_abort(smc, ini->first_contact_local); 1490 mutex_unlock(&smc_server_lgr_pending); 1491 smc->connect_nonblock = 0; 1492 1493 return rc; 1494 } 1495 1496 /* check if received accept type and version matches a proposed one */ 1497 static int smc_connect_check_aclc(struct smc_init_info *ini, 1498 struct smc_clc_msg_accept_confirm *aclc) 1499 { 1500 if (aclc->hdr.version >= SMC_V2) { 1501 if ((aclc->hdr.typev1 == SMC_TYPE_R && 1502 !smcr_indicated(ini->smc_type_v2)) || 1503 (aclc->hdr.typev1 == SMC_TYPE_D && 1504 !smcd_indicated(ini->smc_type_v2))) 1505 return SMC_CLC_DECL_MODEUNSUPP; 1506 } else { 1507 if ((aclc->hdr.typev1 == SMC_TYPE_R && 1508 !smcr_indicated(ini->smc_type_v1)) || 1509 (aclc->hdr.typev1 == SMC_TYPE_D && 1510 !smcd_indicated(ini->smc_type_v1))) 1511 return SMC_CLC_DECL_MODEUNSUPP; 1512 } 1513 1514 return 0; 1515 } 1516 1517 /* perform steps before actually connecting */ 1518 static int __smc_connect(struct smc_sock *smc) 1519 { 1520 u8 version = smc_ism_is_v2_capable() ? SMC_V2 : SMC_V1; 1521 struct smc_clc_msg_accept_confirm *aclc; 1522 struct smc_init_info *ini = NULL; 1523 u8 *buf = NULL; 1524 int rc = 0; 1525 1526 if (smc->use_fallback) 1527 return smc_connect_fallback(smc, smc->fallback_rsn); 1528 1529 /* if peer has not signalled SMC-capability, fall back */ 1530 if (!tcp_sk(smc->clcsock->sk)->syn_smc) 1531 return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC); 1532 1533 /* IPSec connections opt out of SMC optimizations */ 1534 if (using_ipsec(smc)) 1535 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC, 1536 version); 1537 1538 ini = kzalloc_obj(*ini); 1539 if (!ini) 1540 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_MEM, 1541 version); 1542 1543 ini->smcd_version = SMC_V1 | SMC_V2; 1544 ini->smcr_version = SMC_V1 | SMC_V2; 1545 ini->smc_type_v1 = SMC_TYPE_B; 1546 ini->smc_type_v2 = SMC_TYPE_B; 1547 1548 /* get vlan id from IP device */ 1549 if (smc_vlan_by_tcpsk(smc->clcsock, ini)) { 1550 ini->smcd_version &= ~SMC_V1; 1551 ini->smcr_version = 0; 1552 ini->smc_type_v1 = SMC_TYPE_N; 1553 } 1554 1555 rc = smc_find_proposal_devices(smc, ini); 1556 if (rc) 1557 goto fallback; 1558 1559 buf = kzalloc(SMC_CLC_MAX_ACCEPT_LEN, GFP_KERNEL); 1560 if (!buf) { 1561 rc = SMC_CLC_DECL_MEM; 1562 goto fallback; 1563 } 1564 aclc = (struct smc_clc_msg_accept_confirm *)buf; 1565 1566 /* perform CLC handshake */ 1567 rc = smc_connect_clc(smc, aclc, ini); 1568 if (rc) { 1569 /* -EAGAIN on timeout, see tcp_recvmsg() */ 1570 if (rc == -EAGAIN) { 1571 rc = -ETIMEDOUT; 1572 smc->sk.sk_err = ETIMEDOUT; 1573 } 1574 goto vlan_cleanup; 1575 } 1576 1577 /* check if smc modes and versions of CLC proposal and accept match */ 1578 rc = smc_connect_check_aclc(ini, aclc); 1579 version = aclc->hdr.version == SMC_V1 ? SMC_V1 : SMC_V2; 1580 if (rc) 1581 goto vlan_cleanup; 1582 1583 /* depending on previous steps, connect using rdma or ism */ 1584 if (aclc->hdr.typev1 == SMC_TYPE_R) { 1585 ini->smcr_version = version; 1586 rc = smc_connect_rdma(smc, aclc, ini); 1587 } else if (aclc->hdr.typev1 == SMC_TYPE_D) { 1588 ini->smcd_version = version; 1589 rc = smc_connect_ism(smc, aclc, ini); 1590 } 1591 if (rc) 1592 goto vlan_cleanup; 1593 1594 SMC_STAT_CLNT_SUCC_INC(sock_net(smc->clcsock->sk), aclc); 1595 smc_connect_ism_vlan_cleanup(ini); 1596 kfree(buf); 1597 kfree(ini); 1598 return 0; 1599 1600 vlan_cleanup: 1601 smc_connect_ism_vlan_cleanup(ini); 1602 kfree(buf); 1603 fallback: 1604 kfree(ini); 1605 return smc_connect_decline_fallback(smc, rc, version); 1606 } 1607 1608 static void smc_connect_work(struct work_struct *work) 1609 { 1610 struct smc_sock *smc = container_of(work, struct smc_sock, 1611 connect_work); 1612 long timeo = READ_ONCE(smc->sk.sk_sndtimeo); 1613 int rc = 0; 1614 1615 if (!timeo) 1616 timeo = MAX_SCHEDULE_TIMEOUT; 1617 lock_sock(smc->clcsock->sk); 1618 if (smc->clcsock->sk->sk_err) { 1619 smc->sk.sk_err = smc->clcsock->sk->sk_err; 1620 } else if ((1 << smc->clcsock->sk->sk_state) & 1621 (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1622 rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo); 1623 if ((rc == -EPIPE) && 1624 ((1 << smc->clcsock->sk->sk_state) & 1625 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))) 1626 rc = 0; 1627 } 1628 release_sock(smc->clcsock->sk); 1629 lock_sock(&smc->sk); 1630 if (rc != 0 || smc->sk.sk_err) { 1631 smc->sk.sk_state = SMC_CLOSED; 1632 if (!smc->sk.sk_err) 1633 smc->sk.sk_err = (rc == -EAGAIN) ? EPIPE : -rc; 1634 sock_put(&smc->sk); /* passive closing */ 1635 goto out; 1636 } 1637 1638 rc = __smc_connect(smc); 1639 if (rc < 0) 1640 smc->sk.sk_err = -rc; 1641 1642 out: 1643 if (!sock_flag(&smc->sk, SOCK_DEAD)) { 1644 if (smc->sk.sk_err) { 1645 smc->sk.sk_state_change(&smc->sk); 1646 } else { /* allow polling before and after fallback decision */ 1647 smc->clcsock->sk->sk_write_space(smc->clcsock->sk); 1648 smc->sk.sk_write_space(&smc->sk); 1649 } 1650 } 1651 release_sock(&smc->sk); 1652 } 1653 1654 int smc_connect(struct socket *sock, struct sockaddr_unsized *addr, 1655 int alen, int flags) 1656 { 1657 struct sock *sk = sock->sk; 1658 struct smc_sock *smc; 1659 int rc = -EINVAL; 1660 1661 smc = smc_sk(sk); 1662 1663 /* separate smc parameter checking to be safe */ 1664 if (alen < sizeof(addr->sa_family)) 1665 goto out_err; 1666 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6) 1667 goto out_err; 1668 1669 lock_sock(sk); 1670 switch (sock->state) { 1671 default: 1672 rc = -EINVAL; 1673 goto out; 1674 case SS_CONNECTED: 1675 rc = sk->sk_state == SMC_ACTIVE ? -EISCONN : -EINVAL; 1676 goto out; 1677 case SS_CONNECTING: 1678 if (sk->sk_state == SMC_ACTIVE) 1679 goto connected; 1680 break; 1681 case SS_UNCONNECTED: 1682 sock->state = SS_CONNECTING; 1683 break; 1684 } 1685 1686 switch (sk->sk_state) { 1687 default: 1688 goto out; 1689 case SMC_CLOSED: 1690 rc = sock_error(sk) ? : -ECONNABORTED; 1691 sock->state = SS_UNCONNECTED; 1692 goto out; 1693 case SMC_ACTIVE: 1694 rc = -EISCONN; 1695 goto out; 1696 case SMC_INIT: 1697 break; 1698 } 1699 1700 smc_copy_sock_settings_to_clc(smc); 1701 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 1702 if (smc->connect_nonblock) { 1703 rc = -EALREADY; 1704 goto out; 1705 } 1706 rc = kernel_connect(smc->clcsock, (struct sockaddr_unsized *)addr, alen, flags); 1707 if (rc && rc != -EINPROGRESS) 1708 goto out; 1709 1710 if (smc->use_fallback) { 1711 sock->state = rc ? SS_CONNECTING : SS_CONNECTED; 1712 goto out; 1713 } 1714 sock_hold(&smc->sk); /* sock put in passive closing */ 1715 if (flags & O_NONBLOCK) { 1716 if (queue_work(smc_hs_wq, &smc->connect_work)) 1717 smc->connect_nonblock = 1; 1718 rc = -EINPROGRESS; 1719 goto out; 1720 } else { 1721 rc = __smc_connect(smc); 1722 if (rc < 0) 1723 goto out; 1724 } 1725 1726 connected: 1727 rc = 0; 1728 sock->state = SS_CONNECTED; 1729 out: 1730 release_sock(sk); 1731 out_err: 1732 return rc; 1733 } 1734 1735 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc) 1736 { 1737 struct socket *new_clcsock = NULL; 1738 struct sock *lsk = &lsmc->sk; 1739 struct sock *new_sk; 1740 int rc = -EINVAL; 1741 1742 release_sock(lsk); 1743 new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol); 1744 if (!new_sk) { 1745 rc = -ENOMEM; 1746 lsk->sk_err = ENOMEM; 1747 *new_smc = NULL; 1748 lock_sock(lsk); 1749 goto out; 1750 } 1751 *new_smc = smc_sk(new_sk); 1752 1753 mutex_lock(&lsmc->clcsock_release_lock); 1754 if (lsmc->clcsock) 1755 rc = kernel_accept(lsmc->clcsock, &new_clcsock, SOCK_NONBLOCK); 1756 mutex_unlock(&lsmc->clcsock_release_lock); 1757 lock_sock(lsk); 1758 if (rc < 0 && rc != -EAGAIN) 1759 lsk->sk_err = -rc; 1760 if (rc < 0 || lsk->sk_state == SMC_CLOSED) { 1761 new_sk->sk_prot->unhash(new_sk); 1762 if (new_clcsock) 1763 sock_release(new_clcsock); 1764 new_sk->sk_state = SMC_CLOSED; 1765 smc_sock_set_flag(new_sk, SOCK_DEAD); 1766 sock_put(new_sk); /* final */ 1767 *new_smc = NULL; 1768 goto out; 1769 } 1770 1771 /* new clcsock has inherited the smc listen-specific sk_data_ready 1772 * function; switch it back to the original sk_data_ready function 1773 */ 1774 new_clcsock->sk->sk_data_ready = lsmc->clcsk_data_ready; 1775 1776 /* if new clcsock has also inherited the fallback-specific callback 1777 * functions, switch them back to the original ones. 1778 */ 1779 if (lsmc->use_fallback) { 1780 if (lsmc->clcsk_state_change) 1781 new_clcsock->sk->sk_state_change = lsmc->clcsk_state_change; 1782 if (lsmc->clcsk_write_space) 1783 new_clcsock->sk->sk_write_space = lsmc->clcsk_write_space; 1784 if (lsmc->clcsk_error_report) 1785 new_clcsock->sk->sk_error_report = lsmc->clcsk_error_report; 1786 } 1787 1788 (*new_smc)->clcsock = new_clcsock; 1789 out: 1790 return rc; 1791 } 1792 1793 /* add a just created sock to the accept queue of the listen sock as 1794 * candidate for a following socket accept call from user space 1795 */ 1796 static void smc_accept_enqueue(struct sock *parent, struct sock *sk) 1797 { 1798 struct smc_sock *par = smc_sk(parent); 1799 1800 sock_hold(sk); /* sock_put in smc_accept_unlink () */ 1801 spin_lock(&par->accept_q_lock); 1802 list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q); 1803 spin_unlock(&par->accept_q_lock); 1804 sk_acceptq_added(parent); 1805 } 1806 1807 /* remove a socket from the accept queue of its parental listening socket */ 1808 static void smc_accept_unlink(struct sock *sk) 1809 { 1810 struct smc_sock *par = smc_sk(sk)->listen_smc; 1811 1812 spin_lock(&par->accept_q_lock); 1813 list_del_init(&smc_sk(sk)->accept_q); 1814 spin_unlock(&par->accept_q_lock); 1815 sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk); 1816 sock_put(sk); /* sock_hold in smc_accept_enqueue */ 1817 } 1818 1819 /* remove a sock from the accept queue to bind it to a new socket created 1820 * for a socket accept call from user space 1821 */ 1822 struct sock *smc_accept_dequeue(struct sock *parent, 1823 struct socket *new_sock) 1824 { 1825 struct smc_sock *isk, *n; 1826 struct sock *new_sk; 1827 1828 list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) { 1829 new_sk = (struct sock *)isk; 1830 1831 smc_accept_unlink(new_sk); 1832 if (new_sk->sk_state == SMC_CLOSED) { 1833 new_sk->sk_prot->unhash(new_sk); 1834 if (isk->clcsock) { 1835 sock_release(isk->clcsock); 1836 isk->clcsock = NULL; 1837 } 1838 sock_put(new_sk); /* final */ 1839 continue; 1840 } 1841 if (new_sock) { 1842 sock_graft(new_sk, new_sock); 1843 new_sock->state = SS_CONNECTED; 1844 if (isk->use_fallback) { 1845 smc_sk(new_sk)->clcsock->file = new_sock->file; 1846 isk->clcsock->file->private_data = isk->clcsock; 1847 } 1848 } 1849 return new_sk; 1850 } 1851 return NULL; 1852 } 1853 1854 /* clean up for a created but never accepted sock */ 1855 void smc_close_non_accepted(struct sock *sk) 1856 { 1857 struct smc_sock *smc = smc_sk(sk); 1858 1859 sock_hold(sk); /* sock_put below */ 1860 lock_sock(sk); 1861 if (!sk->sk_lingertime) 1862 /* wait for peer closing */ 1863 WRITE_ONCE(sk->sk_lingertime, SMC_MAX_STREAM_WAIT_TIMEOUT); 1864 __smc_release(smc); 1865 release_sock(sk); 1866 sock_put(sk); /* sock_hold above */ 1867 sock_put(sk); /* final sock_put */ 1868 } 1869 1870 static int smcr_serv_conf_first_link(struct smc_sock *smc) 1871 { 1872 struct smc_link *link = smc->conn.lnk; 1873 struct smc_llc_qentry *qentry; 1874 int rc; 1875 1876 /* reg the sndbuf if it was vzalloced*/ 1877 if (smc->conn.sndbuf_desc->is_vm) { 1878 if (smcr_link_reg_buf(link, smc->conn.sndbuf_desc)) 1879 return SMC_CLC_DECL_ERR_REGBUF; 1880 } 1881 1882 /* reg the rmb */ 1883 if (smcr_link_reg_buf(link, smc->conn.rmb_desc)) 1884 return SMC_CLC_DECL_ERR_REGBUF; 1885 1886 /* send CONFIRM LINK request to client over the RoCE fabric */ 1887 rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ); 1888 if (rc < 0) 1889 return SMC_CLC_DECL_TIMEOUT_CL; 1890 1891 /* receive CONFIRM LINK response from client over the RoCE fabric */ 1892 qentry = smc_llc_wait(link->lgr, link, SMC_LLC_WAIT_TIME, 1893 SMC_LLC_CONFIRM_LINK); 1894 if (!qentry) { 1895 struct smc_clc_msg_decline dclc; 1896 1897 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 1898 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 1899 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 1900 } 1901 smc_llc_save_peer_uid(qentry); 1902 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_RESP); 1903 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl); 1904 if (rc) 1905 return SMC_CLC_DECL_RMBE_EC; 1906 1907 /* confirm_rkey is implicit on 1st contact */ 1908 smc->conn.rmb_desc->is_conf_rkey = true; 1909 1910 smc_llc_link_active(link); 1911 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE); 1912 1913 if (link->lgr->max_links > 1) { 1914 down_write(&link->lgr->llc_conf_mutex); 1915 /* initial contact - try to establish second link */ 1916 smc_llc_srv_add_link(link, NULL); 1917 up_write(&link->lgr->llc_conf_mutex); 1918 } 1919 return 0; 1920 } 1921 1922 /* listen worker: finish */ 1923 static void smc_listen_out(struct smc_sock *new_smc) 1924 { 1925 struct smc_sock *lsmc = new_smc->listen_smc; 1926 struct sock *newsmcsk = &new_smc->sk; 1927 1928 if (tcp_sk(new_smc->clcsock->sk)->syn_smc) 1929 atomic_dec(&lsmc->queued_smc_hs); 1930 1931 release_sock(newsmcsk); /* lock in smc_listen_work() */ 1932 if (lsmc->sk.sk_state == SMC_LISTEN) { 1933 lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING); 1934 smc_accept_enqueue(&lsmc->sk, newsmcsk); 1935 release_sock(&lsmc->sk); 1936 } else { /* no longer listening */ 1937 smc_close_non_accepted(newsmcsk); 1938 } 1939 1940 /* Wake up accept */ 1941 lsmc->sk.sk_data_ready(&lsmc->sk); 1942 sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */ 1943 } 1944 1945 /* listen worker: finish in state connected */ 1946 static void smc_listen_out_connected(struct smc_sock *new_smc) 1947 { 1948 struct sock *newsmcsk = &new_smc->sk; 1949 1950 if (newsmcsk->sk_state == SMC_INIT) 1951 newsmcsk->sk_state = SMC_ACTIVE; 1952 1953 smc_listen_out(new_smc); 1954 } 1955 1956 /* listen worker: finish in error state */ 1957 static void smc_listen_out_err(struct smc_sock *new_smc) 1958 { 1959 struct sock *newsmcsk = &new_smc->sk; 1960 struct net *net = sock_net(newsmcsk); 1961 1962 this_cpu_inc(net->smc.smc_stats->srv_hshake_err_cnt); 1963 if (newsmcsk->sk_state == SMC_INIT) 1964 sock_put(&new_smc->sk); /* passive closing */ 1965 newsmcsk->sk_state = SMC_CLOSED; 1966 1967 smc_listen_out(new_smc); 1968 } 1969 1970 /* listen worker: decline and fall back if possible */ 1971 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code, 1972 int local_first, u8 version) 1973 { 1974 /* RDMA setup failed, switch back to TCP */ 1975 smc_conn_abort(new_smc, local_first); 1976 if (reason_code < 0 || 1977 smc_switch_to_fallback(new_smc, reason_code)) { 1978 /* error, no fallback possible */ 1979 smc_listen_out_err(new_smc); 1980 return; 1981 } 1982 if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) { 1983 if (smc_clc_send_decline(new_smc, reason_code, version) < 0) { 1984 smc_listen_out_err(new_smc); 1985 return; 1986 } 1987 } 1988 smc_listen_out_connected(new_smc); 1989 } 1990 1991 /* listen worker: version checking */ 1992 static int smc_listen_v2_check(struct smc_sock *new_smc, 1993 struct smc_clc_msg_proposal *pclc, 1994 struct smc_init_info *ini) 1995 { 1996 struct smc_clc_smcd_v2_extension *pclc_smcd_v2_ext; 1997 struct smc_clc_v2_extension *pclc_v2_ext; 1998 int rc = SMC_CLC_DECL_PEERNOSMC; 1999 2000 ini->smc_type_v1 = pclc->hdr.typev1; 2001 ini->smc_type_v2 = pclc->hdr.typev2; 2002 ini->smcd_version = smcd_indicated(ini->smc_type_v1) ? SMC_V1 : 0; 2003 ini->smcr_version = smcr_indicated(ini->smc_type_v1) ? SMC_V1 : 0; 2004 if (pclc->hdr.version > SMC_V1) { 2005 if (smcd_indicated(ini->smc_type_v2)) 2006 ini->smcd_version |= SMC_V2; 2007 if (smcr_indicated(ini->smc_type_v2)) 2008 ini->smcr_version |= SMC_V2; 2009 } 2010 if (!(ini->smcd_version & SMC_V2) && !(ini->smcr_version & SMC_V2)) { 2011 rc = SMC_CLC_DECL_PEERNOSMC; 2012 goto out; 2013 } 2014 pclc_v2_ext = smc_get_clc_v2_ext(pclc); 2015 if (!pclc_v2_ext) { 2016 ini->smcd_version &= ~SMC_V2; 2017 ini->smcr_version &= ~SMC_V2; 2018 rc = SMC_CLC_DECL_NOV2EXT; 2019 goto out; 2020 } 2021 pclc_smcd_v2_ext = smc_get_clc_smcd_v2_ext(pclc_v2_ext); 2022 if (ini->smcd_version & SMC_V2) { 2023 if (!smc_ism_is_v2_capable()) { 2024 ini->smcd_version &= ~SMC_V2; 2025 rc = SMC_CLC_DECL_NOISM2SUPP; 2026 } else if (!pclc_smcd_v2_ext) { 2027 ini->smcd_version &= ~SMC_V2; 2028 rc = SMC_CLC_DECL_NOV2DEXT; 2029 } else if (!pclc_v2_ext->hdr.eid_cnt && 2030 !pclc_v2_ext->hdr.flag.seid) { 2031 ini->smcd_version &= ~SMC_V2; 2032 rc = SMC_CLC_DECL_NOUEID; 2033 } 2034 } 2035 if (ini->smcr_version & SMC_V2) { 2036 if (!pclc_v2_ext->hdr.eid_cnt) { 2037 ini->smcr_version &= ~SMC_V2; 2038 rc = SMC_CLC_DECL_NOUEID; 2039 } 2040 } 2041 2042 ini->release_nr = pclc_v2_ext->hdr.flag.release; 2043 if (pclc_v2_ext->hdr.flag.release > SMC_RELEASE) 2044 ini->release_nr = SMC_RELEASE; 2045 2046 out: 2047 if (!ini->smcd_version && !ini->smcr_version) 2048 return rc; 2049 2050 return 0; 2051 } 2052 2053 /* listen worker: check prefixes */ 2054 static int smc_listen_prfx_check(struct smc_sock *new_smc, 2055 struct smc_clc_msg_proposal *pclc) 2056 { 2057 struct smc_clc_msg_proposal_prefix *pclc_prfx; 2058 struct socket *newclcsock = new_smc->clcsock; 2059 2060 if (pclc->hdr.typev1 == SMC_TYPE_N) 2061 return 0; 2062 pclc_prfx = smc_clc_proposal_get_prefix(pclc); 2063 if (!pclc_prfx) 2064 return -EPROTO; 2065 if (smc_clc_prfx_match(newclcsock, pclc_prfx)) 2066 return SMC_CLC_DECL_DIFFPREFIX; 2067 2068 return 0; 2069 } 2070 2071 /* listen worker: initialize connection and buffers */ 2072 static int smc_listen_rdma_init(struct smc_sock *new_smc, 2073 struct smc_init_info *ini) 2074 { 2075 int rc; 2076 2077 /* allocate connection / link group */ 2078 rc = smc_conn_create(new_smc, ini); 2079 if (rc) 2080 return rc; 2081 2082 /* create send buffer and rmb */ 2083 if (smc_buf_create(new_smc, false)) { 2084 smc_conn_abort(new_smc, ini->first_contact_local); 2085 return SMC_CLC_DECL_MEM; 2086 } 2087 2088 return 0; 2089 } 2090 2091 /* listen worker: initialize connection and buffers for SMC-D */ 2092 static int smc_listen_ism_init(struct smc_sock *new_smc, 2093 struct smc_init_info *ini) 2094 { 2095 int rc; 2096 2097 rc = smc_conn_create(new_smc, ini); 2098 if (rc) 2099 return rc; 2100 2101 /* Create send and receive buffers */ 2102 rc = smc_buf_create(new_smc, true); 2103 if (rc) { 2104 smc_conn_abort(new_smc, ini->first_contact_local); 2105 return (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : 2106 SMC_CLC_DECL_MEM; 2107 } 2108 2109 return 0; 2110 } 2111 2112 static bool smc_is_already_selected(struct smcd_dev *smcd, 2113 struct smc_init_info *ini, 2114 int matches) 2115 { 2116 int i; 2117 2118 for (i = 0; i < matches; i++) 2119 if (smcd == ini->ism_dev[i]) 2120 return true; 2121 2122 return false; 2123 } 2124 2125 /* check for ISM devices matching proposed ISM devices */ 2126 static void smc_check_ism_v2_match(struct smc_init_info *ini, 2127 u16 proposed_chid, 2128 struct smcd_gid *proposed_gid, 2129 unsigned int *matches) 2130 { 2131 struct smcd_dev *smcd; 2132 2133 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 2134 if (smcd->going_away) 2135 continue; 2136 if (smc_is_already_selected(smcd, ini, *matches)) 2137 continue; 2138 if (smc_ism_get_chid(smcd) == proposed_chid && 2139 !smc_ism_cantalk(proposed_gid, ISM_RESERVED_VLANID, smcd)) { 2140 ini->ism_peer_gid[*matches].gid = proposed_gid->gid; 2141 if (__smc_ism_is_emulated(proposed_chid)) 2142 ini->ism_peer_gid[*matches].gid_ext = 2143 proposed_gid->gid_ext; 2144 /* non-Emulated-ISM's peer gid_ext remains 0. */ 2145 ini->ism_dev[*matches] = smcd; 2146 (*matches)++; 2147 break; 2148 } 2149 } 2150 } 2151 2152 static void smc_init_info_store_rc(u32 rc, struct smc_init_info *ini) 2153 { 2154 if (!ini->rc) 2155 ini->rc = rc; 2156 } 2157 2158 static void smc_find_ism_v2_device_serv(struct smc_sock *new_smc, 2159 struct smc_clc_msg_proposal *pclc, 2160 struct smc_init_info *ini) 2161 { 2162 struct smc_clc_smcd_v2_extension *smcd_v2_ext; 2163 struct smc_clc_v2_extension *smc_v2_ext; 2164 struct smc_clc_msg_smcd *pclc_smcd; 2165 unsigned int matches = 0; 2166 struct smcd_gid smcd_gid; 2167 u8 smcd_version; 2168 u8 *eid = NULL; 2169 int i, rc; 2170 u16 chid; 2171 2172 if (!(ini->smcd_version & SMC_V2) || !smcd_indicated(ini->smc_type_v2)) 2173 goto not_found; 2174 2175 pclc_smcd = smc_get_clc_msg_smcd(pclc); 2176 smc_v2_ext = smc_get_clc_v2_ext(pclc); 2177 smcd_v2_ext = smc_get_clc_smcd_v2_ext(smc_v2_ext); 2178 if (!pclc_smcd || !smc_v2_ext || !smcd_v2_ext) 2179 goto not_found; 2180 2181 mutex_lock(&smcd_dev_list.mutex); 2182 if (pclc_smcd->ism.chid) { 2183 /* check for ISM device matching proposed native ISM device */ 2184 smcd_gid.gid = ntohll(pclc_smcd->ism.gid); 2185 smcd_gid.gid_ext = 0; 2186 smc_check_ism_v2_match(ini, ntohs(pclc_smcd->ism.chid), 2187 &smcd_gid, &matches); 2188 } 2189 for (i = 0; i < smc_v2_ext->hdr.ism_gid_cnt; i++) { 2190 /* check for ISM devices matching proposed non-native ISM 2191 * devices 2192 */ 2193 smcd_gid.gid = ntohll(smcd_v2_ext->gidchid[i].gid); 2194 smcd_gid.gid_ext = 0; 2195 chid = ntohs(smcd_v2_ext->gidchid[i].chid); 2196 if (__smc_ism_is_emulated(chid)) { 2197 if ((i + 1) == smc_v2_ext->hdr.ism_gid_cnt || 2198 chid != ntohs(smcd_v2_ext->gidchid[i + 1].chid)) 2199 /* each Emulated-ISM device takes two GID-CHID 2200 * entries and CHID of the second entry repeats 2201 * that of the first entry. 2202 * 2203 * So check if the next GID-CHID entry exists 2204 * and both two entries' CHIDs are the same. 2205 */ 2206 continue; 2207 smcd_gid.gid_ext = 2208 ntohll(smcd_v2_ext->gidchid[++i].gid); 2209 } 2210 smc_check_ism_v2_match(ini, chid, &smcd_gid, &matches); 2211 } 2212 mutex_unlock(&smcd_dev_list.mutex); 2213 2214 if (!ini->ism_dev[0]) { 2215 smc_init_info_store_rc(SMC_CLC_DECL_NOSMCD2DEV, ini); 2216 goto not_found; 2217 } 2218 2219 smc_ism_get_system_eid(&eid); 2220 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, 2221 smcd_v2_ext->system_eid, eid)) 2222 goto not_found; 2223 2224 /* separate - outside the smcd_dev_list.lock */ 2225 smcd_version = ini->smcd_version; 2226 for (i = 0; i < matches; i++) { 2227 ini->smcd_version = SMC_V2; 2228 ini->is_smcd = true; 2229 ini->ism_selected = i; 2230 rc = smc_listen_ism_init(new_smc, ini); 2231 if (rc) { 2232 smc_init_info_store_rc(rc, ini); 2233 /* try next active ISM device */ 2234 continue; 2235 } 2236 return; /* matching and usable V2 ISM device found */ 2237 } 2238 /* no V2 ISM device could be initialized */ 2239 ini->smcd_version = smcd_version; /* restore original value */ 2240 ini->negotiated_eid[0] = 0; 2241 2242 not_found: 2243 ini->smcd_version &= ~SMC_V2; 2244 ini->ism_dev[0] = NULL; 2245 ini->is_smcd = false; 2246 } 2247 2248 static void smc_find_ism_v1_device_serv(struct smc_sock *new_smc, 2249 struct smc_clc_msg_proposal *pclc, 2250 struct smc_init_info *ini) 2251 { 2252 struct smc_clc_msg_smcd *pclc_smcd = smc_get_clc_msg_smcd(pclc); 2253 int rc = 0; 2254 2255 /* check if ISM V1 is available */ 2256 if (!(ini->smcd_version & SMC_V1) || 2257 !smcd_indicated(ini->smc_type_v1) || 2258 !pclc_smcd) 2259 goto not_found; 2260 ini->is_smcd = true; /* prepare ISM check */ 2261 ini->ism_peer_gid[0].gid = ntohll(pclc_smcd->ism.gid); 2262 ini->ism_peer_gid[0].gid_ext = 0; 2263 rc = smc_find_ism_device(new_smc, ini); 2264 if (rc) 2265 goto not_found; 2266 ini->ism_selected = 0; 2267 rc = smc_listen_ism_init(new_smc, ini); 2268 if (!rc) 2269 return; /* V1 ISM device found */ 2270 2271 not_found: 2272 smc_init_info_store_rc(rc, ini); 2273 ini->smcd_version &= ~SMC_V1; 2274 ini->ism_dev[0] = NULL; 2275 ini->is_smcd = false; 2276 } 2277 2278 /* listen worker: register buffers */ 2279 static int smc_listen_rdma_reg(struct smc_sock *new_smc, bool local_first) 2280 { 2281 struct smc_connection *conn = &new_smc->conn; 2282 2283 if (!local_first) { 2284 /* reg sendbufs if they were vzalloced */ 2285 if (conn->sndbuf_desc->is_vm) { 2286 if (smcr_lgr_reg_sndbufs(conn->lnk, 2287 conn->sndbuf_desc)) 2288 return SMC_CLC_DECL_ERR_REGBUF; 2289 } 2290 if (smcr_lgr_reg_rmbs(conn->lnk, conn->rmb_desc)) 2291 return SMC_CLC_DECL_ERR_REGBUF; 2292 } 2293 2294 return 0; 2295 } 2296 2297 static void smc_find_rdma_v2_device_serv(struct smc_sock *new_smc, 2298 struct smc_clc_msg_proposal *pclc, 2299 struct smc_init_info *ini) 2300 { 2301 struct smc_clc_v2_extension *smc_v2_ext; 2302 u8 smcr_version; 2303 int rc; 2304 2305 if (!(ini->smcr_version & SMC_V2) || !smcr_indicated(ini->smc_type_v2)) 2306 goto not_found; 2307 2308 smc_v2_ext = smc_get_clc_v2_ext(pclc); 2309 if (!smc_v2_ext || 2310 !smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, NULL, NULL)) 2311 goto not_found; 2312 2313 /* prepare RDMA check */ 2314 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN); 2315 memcpy(ini->peer_gid, smc_v2_ext->roce, SMC_GID_SIZE); 2316 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN); 2317 ini->check_smcrv2 = true; 2318 ini->smcrv2.clc_sk = new_smc->clcsock->sk; 2319 ini->smcrv2.saddr = new_smc->clcsock->sk->sk_rcv_saddr; 2320 ini->smcrv2.daddr = smc_ib_gid_to_ipv4(smc_v2_ext->roce); 2321 rc = smc_find_rdma_device(new_smc, ini); 2322 if (rc) { 2323 smc_init_info_store_rc(rc, ini); 2324 goto not_found; 2325 } 2326 if (!ini->smcrv2.uses_gateway) 2327 memcpy(ini->smcrv2.nexthop_mac, pclc->lcl.mac, ETH_ALEN); 2328 2329 smcr_version = ini->smcr_version; 2330 ini->smcr_version = SMC_V2; 2331 rc = smc_listen_rdma_init(new_smc, ini); 2332 if (!rc) { 2333 rc = smc_listen_rdma_reg(new_smc, ini->first_contact_local); 2334 if (rc) 2335 smc_conn_abort(new_smc, ini->first_contact_local); 2336 } 2337 if (!rc) 2338 return; 2339 ini->smcr_version = smcr_version; 2340 smc_init_info_store_rc(rc, ini); 2341 2342 not_found: 2343 ini->smcr_version &= ~SMC_V2; 2344 ini->smcrv2.ib_dev_v2 = NULL; 2345 ini->check_smcrv2 = false; 2346 } 2347 2348 static int smc_find_rdma_v1_device_serv(struct smc_sock *new_smc, 2349 struct smc_clc_msg_proposal *pclc, 2350 struct smc_init_info *ini) 2351 { 2352 int rc; 2353 2354 if (!(ini->smcr_version & SMC_V1) || !smcr_indicated(ini->smc_type_v1)) 2355 return SMC_CLC_DECL_NOSMCDEV; 2356 2357 /* prepare RDMA check */ 2358 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN); 2359 memcpy(ini->peer_gid, pclc->lcl.gid, SMC_GID_SIZE); 2360 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN); 2361 rc = smc_find_rdma_device(new_smc, ini); 2362 if (rc) { 2363 /* no RDMA device found */ 2364 return SMC_CLC_DECL_NOSMCDEV; 2365 } 2366 rc = smc_listen_rdma_init(new_smc, ini); 2367 if (rc) 2368 return rc; 2369 return smc_listen_rdma_reg(new_smc, ini->first_contact_local); 2370 } 2371 2372 /* determine the local device matching to proposal */ 2373 static int smc_listen_find_device(struct smc_sock *new_smc, 2374 struct smc_clc_msg_proposal *pclc, 2375 struct smc_init_info *ini) 2376 { 2377 int prfx_rc; 2378 2379 /* check for ISM device matching V2 proposed device */ 2380 smc_find_ism_v2_device_serv(new_smc, pclc, ini); 2381 if (ini->ism_dev[0]) 2382 return 0; 2383 2384 /* check for matching IP prefix and subnet length (V1) */ 2385 prfx_rc = smc_listen_prfx_check(new_smc, pclc); 2386 if (prfx_rc) 2387 smc_init_info_store_rc(prfx_rc, ini); 2388 2389 /* get vlan id from IP device */ 2390 if (smc_vlan_by_tcpsk(new_smc->clcsock, ini)) 2391 return ini->rc ?: SMC_CLC_DECL_GETVLANERR; 2392 2393 /* check for ISM device matching V1 proposed device */ 2394 if (!prfx_rc) 2395 smc_find_ism_v1_device_serv(new_smc, pclc, ini); 2396 if (ini->ism_dev[0]) 2397 return 0; 2398 2399 if (!smcr_indicated(pclc->hdr.typev1) && 2400 !smcr_indicated(pclc->hdr.typev2)) 2401 /* skip RDMA and decline */ 2402 return ini->rc ?: SMC_CLC_DECL_NOSMCDDEV; 2403 2404 /* check if RDMA V2 is available */ 2405 smc_find_rdma_v2_device_serv(new_smc, pclc, ini); 2406 if (ini->smcrv2.ib_dev_v2) 2407 return 0; 2408 2409 /* check if RDMA V1 is available */ 2410 if (!prfx_rc) { 2411 int rc; 2412 2413 rc = smc_find_rdma_v1_device_serv(new_smc, pclc, ini); 2414 smc_init_info_store_rc(rc, ini); 2415 return (!rc) ? 0 : ini->rc; 2416 } 2417 return prfx_rc; 2418 } 2419 2420 /* listen worker: finish RDMA setup */ 2421 static int smc_listen_rdma_finish(struct smc_sock *new_smc, 2422 struct smc_clc_msg_accept_confirm *cclc, 2423 bool local_first, 2424 struct smc_init_info *ini) 2425 { 2426 struct smc_link *link = new_smc->conn.lnk; 2427 int reason_code = 0; 2428 2429 if (local_first) 2430 smc_link_save_peer_info(link, cclc, ini); 2431 2432 if (smc_rmb_rtoken_handling(&new_smc->conn, link, cclc)) 2433 return SMC_CLC_DECL_ERR_RTOK; 2434 2435 if (local_first) { 2436 if (smc_ib_ready_link(link)) 2437 return SMC_CLC_DECL_ERR_RDYLNK; 2438 /* QP confirmation over RoCE fabric */ 2439 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 2440 reason_code = smcr_serv_conf_first_link(new_smc); 2441 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 2442 } 2443 return reason_code; 2444 } 2445 2446 /* setup for connection of server */ 2447 static void smc_listen_work(struct work_struct *work) 2448 { 2449 struct smc_sock *new_smc = container_of(work, struct smc_sock, 2450 smc_listen_work); 2451 struct socket *newclcsock = new_smc->clcsock; 2452 struct smc_clc_msg_accept_confirm *cclc; 2453 struct smc_clc_msg_proposal_area *buf; 2454 struct smc_clc_msg_proposal *pclc; 2455 struct smc_init_info *ini = NULL; 2456 u8 proposal_version = SMC_V1; 2457 u8 accept_version; 2458 int rc = 0; 2459 2460 lock_sock(&new_smc->sk); /* release in smc_listen_out() */ 2461 if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN) 2462 return smc_listen_out_err(new_smc); 2463 2464 if (new_smc->use_fallback) { 2465 smc_listen_out_connected(new_smc); 2466 return; 2467 } 2468 2469 /* check if peer is smc capable */ 2470 if (!tcp_sk(newclcsock->sk)->syn_smc) { 2471 rc = smc_switch_to_fallback(new_smc, SMC_CLC_DECL_PEERNOSMC); 2472 if (rc) 2473 smc_listen_out_err(new_smc); 2474 else 2475 smc_listen_out_connected(new_smc); 2476 return; 2477 } 2478 2479 /* do inband token exchange - 2480 * wait for and receive SMC Proposal CLC message 2481 */ 2482 buf = kzalloc_obj(*buf); 2483 if (!buf) { 2484 rc = SMC_CLC_DECL_MEM; 2485 goto out_decl; 2486 } 2487 pclc = (struct smc_clc_msg_proposal *)buf; 2488 rc = smc_clc_wait_msg(new_smc, pclc, sizeof(*buf), 2489 SMC_CLC_PROPOSAL, CLC_WAIT_TIME); 2490 if (rc) 2491 goto out_decl; 2492 2493 if (pclc->hdr.version > SMC_V1) 2494 proposal_version = SMC_V2; 2495 2496 /* IPSec connections opt out of SMC optimizations */ 2497 if (using_ipsec(new_smc)) { 2498 rc = SMC_CLC_DECL_IPSEC; 2499 goto out_decl; 2500 } 2501 2502 ini = kzalloc_obj(*ini); 2503 if (!ini) { 2504 rc = SMC_CLC_DECL_MEM; 2505 goto out_decl; 2506 } 2507 2508 /* initial version checking */ 2509 rc = smc_listen_v2_check(new_smc, pclc, ini); 2510 if (rc) 2511 goto out_decl; 2512 2513 rc = smc_clc_srv_v2x_features_validate(new_smc, pclc, ini); 2514 if (rc) 2515 goto out_decl; 2516 2517 mutex_lock(&smc_server_lgr_pending); 2518 smc_rx_init(new_smc); 2519 smc_tx_init(new_smc); 2520 2521 /* determine ISM or RoCE device used for connection */ 2522 rc = smc_listen_find_device(new_smc, pclc, ini); 2523 if (rc) 2524 goto out_unlock; 2525 2526 /* send SMC Accept CLC message */ 2527 accept_version = ini->is_smcd ? ini->smcd_version : ini->smcr_version; 2528 rc = smc_clc_send_accept(new_smc, ini->first_contact_local, 2529 accept_version, ini->negotiated_eid, ini); 2530 if (rc) 2531 goto out_unlock; 2532 2533 /* SMC-D does not need this lock any more */ 2534 if (ini->is_smcd) 2535 mutex_unlock(&smc_server_lgr_pending); 2536 2537 /* receive SMC Confirm CLC message */ 2538 memset(buf, 0, sizeof(*buf)); 2539 cclc = (struct smc_clc_msg_accept_confirm *)buf; 2540 rc = smc_clc_wait_msg(new_smc, cclc, sizeof(*buf), 2541 SMC_CLC_CONFIRM, CLC_WAIT_TIME); 2542 if (rc) { 2543 if (!ini->is_smcd) 2544 goto out_unlock; 2545 goto out_decl; 2546 } 2547 2548 rc = smc_clc_v2x_features_confirm_check(cclc, ini); 2549 if (rc) { 2550 if (!ini->is_smcd) 2551 goto out_unlock; 2552 goto out_decl; 2553 } 2554 2555 /* fce smc release version is needed in smc_listen_rdma_finish, 2556 * so save fce info here. 2557 */ 2558 smc_conn_save_peer_info_fce(new_smc, cclc); 2559 2560 /* finish worker */ 2561 if (!ini->is_smcd) { 2562 rc = smc_listen_rdma_finish(new_smc, cclc, 2563 ini->first_contact_local, ini); 2564 if (rc) 2565 goto out_unlock; 2566 mutex_unlock(&smc_server_lgr_pending); 2567 } 2568 smc_conn_save_peer_info(new_smc, cclc); 2569 2570 if (ini->is_smcd && 2571 smc_ism_support_dmb_nocopy(new_smc->conn.lgr->smcd)) { 2572 rc = smcd_buf_attach(new_smc); 2573 if (rc) 2574 goto out_decl; 2575 } 2576 2577 SMC_STAT_SERV_SUCC_INC(sock_net(newclcsock->sk), ini); 2578 /* smc_listen_out() will release smcsk */ 2579 smc_listen_out_connected(new_smc); 2580 goto out_free; 2581 2582 out_unlock: 2583 mutex_unlock(&smc_server_lgr_pending); 2584 out_decl: 2585 smc_listen_decline(new_smc, rc, ini ? ini->first_contact_local : 0, 2586 proposal_version); 2587 out_free: 2588 kfree(ini); 2589 kfree(buf); 2590 } 2591 2592 static void smc_tcp_listen_work(struct work_struct *work) 2593 { 2594 struct smc_sock *lsmc = container_of(work, struct smc_sock, 2595 tcp_listen_work); 2596 struct sock *lsk = &lsmc->sk; 2597 struct smc_sock *new_smc; 2598 int rc = 0; 2599 2600 lock_sock(lsk); 2601 while (lsk->sk_state == SMC_LISTEN) { 2602 rc = smc_clcsock_accept(lsmc, &new_smc); 2603 if (rc) /* clcsock accept queue empty or error */ 2604 goto out; 2605 if (!new_smc) 2606 continue; 2607 2608 if (tcp_sk(new_smc->clcsock->sk)->syn_smc) 2609 atomic_inc(&lsmc->queued_smc_hs); 2610 2611 new_smc->listen_smc = lsmc; 2612 new_smc->use_fallback = lsmc->use_fallback; 2613 new_smc->fallback_rsn = lsmc->fallback_rsn; 2614 sock_hold(lsk); /* sock_put in smc_listen_work */ 2615 INIT_WORK(&new_smc->smc_listen_work, smc_listen_work); 2616 smc_copy_sock_settings_to_smc(new_smc); 2617 sock_hold(&new_smc->sk); /* sock_put in passive closing */ 2618 if (!queue_work(smc_hs_wq, &new_smc->smc_listen_work)) 2619 sock_put(&new_smc->sk); 2620 } 2621 2622 out: 2623 release_sock(lsk); 2624 sock_put(&lsmc->sk); /* sock_hold in smc_clcsock_data_ready() */ 2625 } 2626 2627 static void smc_clcsock_data_ready(struct sock *listen_clcsock) 2628 { 2629 struct smc_sock *lsmc; 2630 2631 read_lock_bh(&listen_clcsock->sk_callback_lock); 2632 lsmc = smc_clcsock_user_data(listen_clcsock); 2633 if (!lsmc) 2634 goto out; 2635 lsmc->clcsk_data_ready(listen_clcsock); 2636 if (lsmc->sk.sk_state == SMC_LISTEN) { 2637 sock_hold(&lsmc->sk); /* sock_put in smc_tcp_listen_work() */ 2638 if (!queue_work(smc_tcp_ls_wq, &lsmc->tcp_listen_work)) 2639 sock_put(&lsmc->sk); 2640 } 2641 out: 2642 read_unlock_bh(&listen_clcsock->sk_callback_lock); 2643 } 2644 2645 int smc_listen(struct socket *sock, int backlog) 2646 { 2647 struct sock *sk = sock->sk; 2648 struct smc_sock *smc; 2649 int rc; 2650 2651 smc = smc_sk(sk); 2652 lock_sock(sk); 2653 2654 rc = -EINVAL; 2655 if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) || 2656 smc->connect_nonblock || sock->state != SS_UNCONNECTED) 2657 goto out; 2658 2659 rc = 0; 2660 if (sk->sk_state == SMC_LISTEN) { 2661 sk->sk_max_ack_backlog = backlog; 2662 goto out; 2663 } 2664 /* some socket options are handled in core, so we could not apply 2665 * them to the clc socket -- copy smc socket options to clc socket 2666 */ 2667 smc_copy_sock_settings_to_clc(smc); 2668 if (!smc->use_fallback) 2669 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 2670 2671 /* save original sk_data_ready function and establish 2672 * smc-specific sk_data_ready function 2673 */ 2674 write_lock_bh(&smc->clcsock->sk->sk_callback_lock); 2675 __rcu_assign_sk_user_data_with_flags(smc->clcsock->sk, smc, 2676 SK_USER_DATA_NOCOPY); 2677 smc_clcsock_replace_cb(&smc->clcsock->sk->sk_data_ready, 2678 smc_clcsock_data_ready, &smc->clcsk_data_ready); 2679 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock); 2680 2681 /* save original ops */ 2682 smc->ori_af_ops = inet_csk(smc->clcsock->sk)->icsk_af_ops; 2683 2684 smc->af_ops = *smc->ori_af_ops; 2685 smc->af_ops.syn_recv_sock = smc_tcp_syn_recv_sock; 2686 2687 inet_csk(smc->clcsock->sk)->icsk_af_ops = &smc->af_ops; 2688 2689 if (smc->limit_smc_hs) 2690 tcp_sk(smc->clcsock->sk)->smc_hs_congested = smc_hs_congested; 2691 2692 rc = kernel_listen(smc->clcsock, backlog); 2693 if (rc) { 2694 write_lock_bh(&smc->clcsock->sk->sk_callback_lock); 2695 smc_clcsock_restore_cb(&smc->clcsock->sk->sk_data_ready, 2696 &smc->clcsk_data_ready); 2697 rcu_assign_sk_user_data(smc->clcsock->sk, NULL); 2698 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock); 2699 goto out; 2700 } 2701 sock_set_flag(sk, SOCK_RCU_FREE); 2702 sk->sk_max_ack_backlog = backlog; 2703 sk->sk_ack_backlog = 0; 2704 sk->sk_state = SMC_LISTEN; 2705 2706 out: 2707 release_sock(sk); 2708 return rc; 2709 } 2710 2711 int smc_accept(struct socket *sock, struct socket *new_sock, 2712 struct proto_accept_arg *arg) 2713 { 2714 struct sock *sk = sock->sk, *nsk; 2715 DECLARE_WAITQUEUE(wait, current); 2716 struct smc_sock *lsmc; 2717 long timeo; 2718 int rc = 0; 2719 2720 lsmc = smc_sk(sk); 2721 sock_hold(sk); /* sock_put below */ 2722 lock_sock(sk); 2723 2724 if (lsmc->sk.sk_state != SMC_LISTEN) { 2725 rc = -EINVAL; 2726 release_sock(sk); 2727 goto out; 2728 } 2729 2730 /* Wait for an incoming connection */ 2731 timeo = sock_rcvtimeo(sk, arg->flags & O_NONBLOCK); 2732 add_wait_queue_exclusive(sk_sleep(sk), &wait); 2733 while (!(nsk = smc_accept_dequeue(sk, new_sock))) { 2734 set_current_state(TASK_INTERRUPTIBLE); 2735 if (!timeo) { 2736 rc = -EAGAIN; 2737 break; 2738 } 2739 release_sock(sk); 2740 timeo = schedule_timeout(timeo); 2741 /* wakeup by sk_data_ready in smc_listen_work() */ 2742 sched_annotate_sleep(); 2743 lock_sock(sk); 2744 if (signal_pending(current)) { 2745 rc = sock_intr_errno(timeo); 2746 break; 2747 } 2748 } 2749 set_current_state(TASK_RUNNING); 2750 remove_wait_queue(sk_sleep(sk), &wait); 2751 2752 if (!rc) 2753 rc = sock_error(nsk); 2754 release_sock(sk); 2755 if (rc) 2756 goto out; 2757 2758 if (lsmc->sockopt_defer_accept && !(arg->flags & O_NONBLOCK)) { 2759 /* wait till data arrives on the socket */ 2760 timeo = secs_to_jiffies(lsmc->sockopt_defer_accept); 2761 if (smc_sk(nsk)->use_fallback) { 2762 struct sock *clcsk = smc_sk(nsk)->clcsock->sk; 2763 2764 lock_sock(clcsk); 2765 if (skb_queue_empty(&clcsk->sk_receive_queue)) 2766 sk_wait_data(clcsk, &timeo, NULL); 2767 release_sock(clcsk); 2768 } else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) { 2769 lock_sock(nsk); 2770 smc_rx_wait(smc_sk(nsk), &timeo, 0, smc_rx_data_available); 2771 release_sock(nsk); 2772 } 2773 } 2774 2775 out: 2776 sock_put(sk); /* sock_hold above */ 2777 return rc; 2778 } 2779 2780 int smc_getname(struct socket *sock, struct sockaddr *addr, 2781 int peer) 2782 { 2783 struct smc_sock *smc; 2784 2785 if (peer && (sock->sk->sk_state != SMC_ACTIVE) && 2786 (sock->sk->sk_state != SMC_APPCLOSEWAIT1)) 2787 return -ENOTCONN; 2788 2789 smc = smc_sk(sock->sk); 2790 2791 return smc->clcsock->ops->getname(smc->clcsock, addr, peer); 2792 } 2793 2794 int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 2795 { 2796 struct sock *sk = sock->sk; 2797 struct smc_sock *smc; 2798 int rc; 2799 2800 smc = smc_sk(sk); 2801 lock_sock(sk); 2802 2803 /* SMC does not support connect with fastopen */ 2804 if (msg->msg_flags & MSG_FASTOPEN) { 2805 /* not connected yet, fallback */ 2806 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 2807 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP); 2808 if (rc) 2809 goto out; 2810 } else { 2811 rc = -EINVAL; 2812 goto out; 2813 } 2814 } else if ((sk->sk_state != SMC_ACTIVE) && 2815 (sk->sk_state != SMC_APPCLOSEWAIT1) && 2816 (sk->sk_state != SMC_INIT)) { 2817 rc = -EPIPE; 2818 goto out; 2819 } 2820 2821 if (smc->use_fallback) { 2822 rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len); 2823 } else { 2824 rc = smc_tx_sendmsg(smc, msg, len); 2825 SMC_STAT_TX_PAYLOAD(smc, len, rc); 2826 } 2827 out: 2828 release_sock(sk); 2829 return rc; 2830 } 2831 2832 int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 2833 int flags) 2834 { 2835 struct sock *sk = sock->sk; 2836 struct smc_sock *smc; 2837 int rc = -ENOTCONN; 2838 2839 smc = smc_sk(sk); 2840 lock_sock(sk); 2841 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 2842 /* socket was connected before, no more data to read */ 2843 rc = 0; 2844 goto out; 2845 } 2846 if ((sk->sk_state == SMC_INIT) || 2847 (sk->sk_state == SMC_LISTEN) || 2848 (sk->sk_state == SMC_CLOSED)) 2849 goto out; 2850 2851 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 2852 rc = 0; 2853 goto out; 2854 } 2855 2856 if (smc->use_fallback) { 2857 rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags); 2858 } else { 2859 msg->msg_namelen = 0; 2860 rc = smc_rx_recvmsg(smc, msg, NULL, len, flags); 2861 SMC_STAT_RX_PAYLOAD(smc, rc, rc); 2862 } 2863 2864 out: 2865 release_sock(sk); 2866 return rc; 2867 } 2868 2869 static __poll_t smc_accept_poll(struct sock *parent) 2870 { 2871 struct smc_sock *isk = smc_sk(parent); 2872 __poll_t mask = 0; 2873 2874 spin_lock(&isk->accept_q_lock); 2875 if (!list_empty(&isk->accept_q)) 2876 mask = EPOLLIN | EPOLLRDNORM; 2877 spin_unlock(&isk->accept_q_lock); 2878 2879 return mask; 2880 } 2881 2882 __poll_t smc_poll(struct file *file, struct socket *sock, 2883 poll_table *wait) 2884 { 2885 struct sock *sk = sock->sk; 2886 struct smc_sock *smc; 2887 __poll_t mask = 0; 2888 2889 if (!sk) 2890 return EPOLLNVAL; 2891 2892 smc = smc_sk(sock->sk); 2893 if (smc->use_fallback) { 2894 /* delegate to CLC child sock */ 2895 mask = smc->clcsock->ops->poll(file, smc->clcsock, wait); 2896 sk->sk_err = smc->clcsock->sk->sk_err; 2897 } else { 2898 if (sk->sk_state != SMC_CLOSED) 2899 sock_poll_wait(file, sock, wait); 2900 if (sk->sk_err) 2901 mask |= EPOLLERR; 2902 if ((sk->sk_shutdown == SHUTDOWN_MASK) || 2903 (sk->sk_state == SMC_CLOSED)) 2904 mask |= EPOLLHUP; 2905 if (sk->sk_state == SMC_LISTEN) { 2906 /* woken up by sk_data_ready in smc_listen_work() */ 2907 mask |= smc_accept_poll(sk); 2908 } else if (smc->use_fallback) { /* as result of connect_work()*/ 2909 mask |= smc->clcsock->ops->poll(file, smc->clcsock, 2910 wait); 2911 sk->sk_err = smc->clcsock->sk->sk_err; 2912 } else { 2913 if ((sk->sk_state != SMC_INIT && 2914 atomic_read(&smc->conn.sndbuf_space)) || 2915 sk->sk_shutdown & SEND_SHUTDOWN) { 2916 mask |= EPOLLOUT | EPOLLWRNORM; 2917 } else { 2918 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2919 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2920 2921 if (sk->sk_state != SMC_INIT) { 2922 /* Race breaker the same way as tcp_poll(). */ 2923 smp_mb__after_atomic(); 2924 if (atomic_read(&smc->conn.sndbuf_space)) 2925 mask |= EPOLLOUT | EPOLLWRNORM; 2926 } 2927 } 2928 if (atomic_read(&smc->conn.bytes_to_rcv)) 2929 mask |= EPOLLIN | EPOLLRDNORM; 2930 if (sk->sk_shutdown & RCV_SHUTDOWN) 2931 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP; 2932 if (sk->sk_state == SMC_APPCLOSEWAIT1) 2933 mask |= EPOLLIN; 2934 if (smc->conn.urg_state == SMC_URG_VALID) 2935 mask |= EPOLLPRI; 2936 } 2937 } 2938 2939 return mask; 2940 } 2941 2942 int smc_shutdown(struct socket *sock, int how) 2943 { 2944 struct sock *sk = sock->sk; 2945 bool do_shutdown = true; 2946 struct smc_sock *smc; 2947 int rc = -EINVAL; 2948 int old_state; 2949 int rc1 = 0; 2950 2951 smc = smc_sk(sk); 2952 2953 if ((how < SHUT_RD) || (how > SHUT_RDWR)) 2954 return rc; 2955 2956 lock_sock(sk); 2957 2958 if (sock->state == SS_CONNECTING) { 2959 if (sk->sk_state == SMC_ACTIVE) 2960 sock->state = SS_CONNECTED; 2961 else if (sk->sk_state == SMC_PEERCLOSEWAIT1 || 2962 sk->sk_state == SMC_PEERCLOSEWAIT2 || 2963 sk->sk_state == SMC_APPCLOSEWAIT1 || 2964 sk->sk_state == SMC_APPCLOSEWAIT2 || 2965 sk->sk_state == SMC_APPFINCLOSEWAIT) 2966 sock->state = SS_DISCONNECTING; 2967 } 2968 2969 rc = -ENOTCONN; 2970 if ((sk->sk_state != SMC_ACTIVE) && 2971 (sk->sk_state != SMC_PEERCLOSEWAIT1) && 2972 (sk->sk_state != SMC_PEERCLOSEWAIT2) && 2973 (sk->sk_state != SMC_APPCLOSEWAIT1) && 2974 (sk->sk_state != SMC_APPCLOSEWAIT2) && 2975 (sk->sk_state != SMC_APPFINCLOSEWAIT)) 2976 goto out; 2977 if (smc->use_fallback) { 2978 rc = kernel_sock_shutdown(smc->clcsock, how); 2979 sk->sk_shutdown = smc->clcsock->sk->sk_shutdown; 2980 if (sk->sk_shutdown == SHUTDOWN_MASK) { 2981 sk->sk_state = SMC_CLOSED; 2982 sk->sk_socket->state = SS_UNCONNECTED; 2983 sock_put(sk); 2984 } 2985 goto out; 2986 } 2987 switch (how) { 2988 case SHUT_RDWR: /* shutdown in both directions */ 2989 old_state = sk->sk_state; 2990 rc = smc_close_active(smc); 2991 if (old_state == SMC_ACTIVE && 2992 sk->sk_state == SMC_PEERCLOSEWAIT1) 2993 do_shutdown = false; 2994 break; 2995 case SHUT_WR: 2996 rc = smc_close_shutdown_write(smc); 2997 break; 2998 case SHUT_RD: 2999 rc = 0; 3000 /* nothing more to do because peer is not involved */ 3001 break; 3002 } 3003 if (do_shutdown && smc->clcsock) 3004 rc1 = kernel_sock_shutdown(smc->clcsock, how); 3005 /* map sock_shutdown_cmd constants to sk_shutdown value range */ 3006 sk->sk_shutdown |= how + 1; 3007 3008 if (sk->sk_state == SMC_CLOSED) 3009 sock->state = SS_UNCONNECTED; 3010 else 3011 sock->state = SS_DISCONNECTING; 3012 out: 3013 release_sock(sk); 3014 return rc ? rc : rc1; 3015 } 3016 3017 static int __smc_getsockopt(struct socket *sock, int level, int optname, 3018 char __user *optval, int __user *optlen) 3019 { 3020 struct smc_sock *smc; 3021 int val, len; 3022 3023 smc = smc_sk(sock->sk); 3024 3025 if (get_user(len, optlen)) 3026 return -EFAULT; 3027 3028 len = min_t(int, len, sizeof(int)); 3029 3030 if (len < 0) 3031 return -EINVAL; 3032 3033 switch (optname) { 3034 case SMC_LIMIT_HS: 3035 val = smc->limit_smc_hs; 3036 break; 3037 default: 3038 return -EOPNOTSUPP; 3039 } 3040 3041 if (put_user(len, optlen)) 3042 return -EFAULT; 3043 if (copy_to_user(optval, &val, len)) 3044 return -EFAULT; 3045 3046 return 0; 3047 } 3048 3049 static int __smc_setsockopt(struct socket *sock, int level, int optname, 3050 sockptr_t optval, unsigned int optlen) 3051 { 3052 struct sock *sk = sock->sk; 3053 struct smc_sock *smc; 3054 int val, rc; 3055 3056 smc = smc_sk(sk); 3057 3058 /* pre-fetch user data outside the lock */ 3059 if (optname == SMC_LIMIT_HS) { 3060 if (optlen < sizeof(int)) 3061 return -EINVAL; 3062 if (copy_from_sockptr(&val, optval, sizeof(int))) 3063 return -EFAULT; 3064 } 3065 3066 lock_sock(sk); 3067 switch (optname) { 3068 case SMC_LIMIT_HS: 3069 smc->limit_smc_hs = !!val; 3070 rc = 0; 3071 break; 3072 default: 3073 rc = -EOPNOTSUPP; 3074 break; 3075 } 3076 release_sock(sk); 3077 3078 return rc; 3079 } 3080 3081 int smc_setsockopt(struct socket *sock, int level, int optname, 3082 sockptr_t optval, unsigned int optlen) 3083 { 3084 struct sock *sk = sock->sk; 3085 struct smc_sock *smc; 3086 int val, rc; 3087 3088 if (level == SOL_TCP && optname == TCP_ULP) 3089 return -EOPNOTSUPP; 3090 else if (level == SOL_SMC) 3091 return __smc_setsockopt(sock, level, optname, optval, optlen); 3092 3093 smc = smc_sk(sk); 3094 3095 /* generic setsockopts reaching us here always apply to the 3096 * CLC socket 3097 */ 3098 mutex_lock(&smc->clcsock_release_lock); 3099 if (!smc->clcsock) { 3100 mutex_unlock(&smc->clcsock_release_lock); 3101 return -EBADF; 3102 } 3103 if (unlikely(!smc->clcsock->ops->setsockopt)) 3104 rc = -EOPNOTSUPP; 3105 else 3106 rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname, 3107 optval, optlen); 3108 if (smc->clcsock->sk->sk_err) { 3109 sk->sk_err = smc->clcsock->sk->sk_err; 3110 sk_error_report(sk); 3111 } 3112 mutex_unlock(&smc->clcsock_release_lock); 3113 3114 if (optlen < sizeof(int)) 3115 return -EINVAL; 3116 if (copy_from_sockptr(&val, optval, sizeof(int))) 3117 return -EFAULT; 3118 3119 lock_sock(sk); 3120 if (rc || smc->use_fallback) 3121 goto out; 3122 switch (optname) { 3123 case TCP_FASTOPEN: 3124 case TCP_FASTOPEN_CONNECT: 3125 case TCP_FASTOPEN_KEY: 3126 case TCP_FASTOPEN_NO_COOKIE: 3127 /* option not supported by SMC */ 3128 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 3129 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP); 3130 } else { 3131 rc = -EINVAL; 3132 } 3133 break; 3134 case TCP_NODELAY: 3135 if (sk->sk_state != SMC_INIT && 3136 sk->sk_state != SMC_LISTEN && 3137 sk->sk_state != SMC_CLOSED) { 3138 if (val) { 3139 SMC_STAT_INC(smc, ndly_cnt); 3140 smc_tx_pending(&smc->conn); 3141 cancel_delayed_work(&smc->conn.tx_work); 3142 } 3143 } 3144 break; 3145 case TCP_CORK: 3146 if (sk->sk_state != SMC_INIT && 3147 sk->sk_state != SMC_LISTEN && 3148 sk->sk_state != SMC_CLOSED) { 3149 if (!val) { 3150 SMC_STAT_INC(smc, cork_cnt); 3151 smc_tx_pending(&smc->conn); 3152 cancel_delayed_work(&smc->conn.tx_work); 3153 } 3154 } 3155 break; 3156 case TCP_DEFER_ACCEPT: 3157 smc->sockopt_defer_accept = val; 3158 break; 3159 default: 3160 break; 3161 } 3162 out: 3163 release_sock(sk); 3164 3165 return rc; 3166 } 3167 3168 int smc_getsockopt(struct socket *sock, int level, int optname, 3169 char __user *optval, int __user *optlen) 3170 { 3171 struct smc_sock *smc; 3172 int rc; 3173 3174 if (level == SOL_SMC) 3175 return __smc_getsockopt(sock, level, optname, optval, optlen); 3176 3177 smc = smc_sk(sock->sk); 3178 mutex_lock(&smc->clcsock_release_lock); 3179 if (!smc->clcsock) { 3180 mutex_unlock(&smc->clcsock_release_lock); 3181 return -EBADF; 3182 } 3183 /* socket options apply to the CLC socket */ 3184 if (unlikely(!smc->clcsock->ops->getsockopt)) { 3185 mutex_unlock(&smc->clcsock_release_lock); 3186 return -EOPNOTSUPP; 3187 } 3188 rc = smc->clcsock->ops->getsockopt(smc->clcsock, level, optname, 3189 optval, optlen); 3190 mutex_unlock(&smc->clcsock_release_lock); 3191 return rc; 3192 } 3193 3194 int smc_ioctl(struct socket *sock, unsigned int cmd, 3195 unsigned long arg) 3196 { 3197 union smc_host_cursor cons, urg; 3198 struct smc_connection *conn; 3199 struct smc_sock *smc; 3200 int answ; 3201 3202 smc = smc_sk(sock->sk); 3203 conn = &smc->conn; 3204 lock_sock(&smc->sk); 3205 if (smc->use_fallback) { 3206 if (!smc->clcsock) { 3207 release_sock(&smc->sk); 3208 return -EBADF; 3209 } 3210 answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg); 3211 release_sock(&smc->sk); 3212 return answ; 3213 } 3214 switch (cmd) { 3215 case SIOCINQ: /* same as FIONREAD */ 3216 if (smc->sk.sk_state == SMC_LISTEN) { 3217 release_sock(&smc->sk); 3218 return -EINVAL; 3219 } 3220 if (smc->sk.sk_state == SMC_INIT || 3221 smc->sk.sk_state == SMC_CLOSED) 3222 answ = 0; 3223 else 3224 answ = atomic_read(&smc->conn.bytes_to_rcv); 3225 break; 3226 case SIOCOUTQ: 3227 /* output queue size (not send + not acked) */ 3228 if (smc->sk.sk_state == SMC_LISTEN) { 3229 release_sock(&smc->sk); 3230 return -EINVAL; 3231 } 3232 if (smc->sk.sk_state == SMC_INIT || 3233 smc->sk.sk_state == SMC_CLOSED) 3234 answ = 0; 3235 else 3236 answ = smc->conn.sndbuf_desc->len - 3237 atomic_read(&smc->conn.sndbuf_space); 3238 break; 3239 case SIOCOUTQNSD: 3240 /* output queue size (not send only) */ 3241 if (smc->sk.sk_state == SMC_LISTEN) { 3242 release_sock(&smc->sk); 3243 return -EINVAL; 3244 } 3245 if (smc->sk.sk_state == SMC_INIT || 3246 smc->sk.sk_state == SMC_CLOSED) 3247 answ = 0; 3248 else 3249 answ = smc_tx_prepared_sends(&smc->conn); 3250 break; 3251 case SIOCATMARK: 3252 if (smc->sk.sk_state == SMC_LISTEN) { 3253 release_sock(&smc->sk); 3254 return -EINVAL; 3255 } 3256 if (smc->sk.sk_state == SMC_INIT || 3257 smc->sk.sk_state == SMC_CLOSED) { 3258 answ = 0; 3259 } else { 3260 smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn); 3261 smc_curs_copy(&urg, &conn->urg_curs, conn); 3262 answ = smc_curs_diff(conn->rmb_desc->len, 3263 &cons, &urg) == 1; 3264 } 3265 break; 3266 default: 3267 release_sock(&smc->sk); 3268 return -ENOIOCTLCMD; 3269 } 3270 release_sock(&smc->sk); 3271 3272 return put_user(answ, (int __user *)arg); 3273 } 3274 3275 /* Map the affected portions of the rmbe into an spd, note the number of bytes 3276 * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor 3277 * updates till whenever a respective page has been fully processed. 3278 * Note that subsequent recv() calls have to wait till all splice() processing 3279 * completed. 3280 */ 3281 ssize_t smc_splice_read(struct socket *sock, loff_t *ppos, 3282 struct pipe_inode_info *pipe, size_t len, 3283 unsigned int flags) 3284 { 3285 struct sock *sk = sock->sk; 3286 struct smc_sock *smc; 3287 int rc = -ENOTCONN; 3288 3289 smc = smc_sk(sk); 3290 lock_sock(sk); 3291 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 3292 /* socket was connected before, no more data to read */ 3293 rc = 0; 3294 goto out; 3295 } 3296 if (sk->sk_state == SMC_INIT || 3297 sk->sk_state == SMC_LISTEN || 3298 sk->sk_state == SMC_CLOSED) 3299 goto out; 3300 3301 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 3302 rc = 0; 3303 goto out; 3304 } 3305 3306 if (smc->use_fallback) { 3307 rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos, 3308 pipe, len, flags); 3309 } else { 3310 if (*ppos) { 3311 rc = -ESPIPE; 3312 goto out; 3313 } 3314 if (flags & SPLICE_F_NONBLOCK) 3315 flags = MSG_DONTWAIT; 3316 else 3317 flags = 0; 3318 SMC_STAT_INC(smc, splice_cnt); 3319 rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags); 3320 } 3321 out: 3322 release_sock(sk); 3323 3324 return rc; 3325 } 3326 3327 /* must look like tcp */ 3328 static const struct proto_ops smc_sock_ops = { 3329 .family = PF_SMC, 3330 .owner = THIS_MODULE, 3331 .release = smc_release, 3332 .bind = smc_bind, 3333 .connect = smc_connect, 3334 .socketpair = sock_no_socketpair, 3335 .accept = smc_accept, 3336 .getname = smc_getname, 3337 .poll = smc_poll, 3338 .ioctl = smc_ioctl, 3339 .listen = smc_listen, 3340 .shutdown = smc_shutdown, 3341 .setsockopt = smc_setsockopt, 3342 .getsockopt = smc_getsockopt, 3343 .sendmsg = smc_sendmsg, 3344 .recvmsg = smc_recvmsg, 3345 .mmap = sock_no_mmap, 3346 .splice_read = smc_splice_read, 3347 }; 3348 3349 int smc_create_clcsk(struct net *net, struct sock *sk, int family) 3350 { 3351 struct smc_sock *smc = smc_sk(sk); 3352 int rc; 3353 3354 rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, 3355 &smc->clcsock); 3356 if (rc) 3357 return rc; 3358 3359 /* smc_clcsock_release() does not wait smc->clcsock->sk's 3360 * destruction; its sk_state might not be TCP_CLOSE after 3361 * smc->sk is close()d, and TCP timers can be fired later, 3362 * which need net ref. 3363 */ 3364 sk = smc->clcsock->sk; 3365 sk_net_refcnt_upgrade(sk); 3366 return 0; 3367 } 3368 3369 static int smc_create(struct net *net, struct socket *sock, int protocol, 3370 int kern) 3371 { 3372 int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET; 3373 struct sock *sk; 3374 int rc; 3375 3376 rc = -ESOCKTNOSUPPORT; 3377 if (sock->type != SOCK_STREAM) 3378 goto out; 3379 3380 rc = -EPROTONOSUPPORT; 3381 if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6) 3382 goto out; 3383 3384 rc = -ENOBUFS; 3385 sock->ops = &smc_sock_ops; 3386 sock->state = SS_UNCONNECTED; 3387 sk = smc_sock_alloc(net, sock, protocol); 3388 if (!sk) 3389 goto out; 3390 3391 rc = smc_create_clcsk(net, sk, family); 3392 if (rc) { 3393 sk_common_release(sk); 3394 sock->sk = NULL; 3395 } 3396 out: 3397 return rc; 3398 } 3399 3400 static const struct net_proto_family smc_sock_family_ops = { 3401 .family = PF_SMC, 3402 .owner = THIS_MODULE, 3403 .create = smc_create, 3404 }; 3405 3406 unsigned int smc_net_id; 3407 3408 static __net_init int smc_net_init(struct net *net) 3409 { 3410 int rc; 3411 3412 rc = smc_sysctl_net_init(net); 3413 if (rc) 3414 return rc; 3415 return smc_pnet_net_init(net); 3416 } 3417 3418 static void __net_exit smc_net_exit(struct net *net) 3419 { 3420 smc_sysctl_net_exit(net); 3421 smc_pnet_net_exit(net); 3422 } 3423 3424 static __net_init int smc_net_stat_init(struct net *net) 3425 { 3426 return smc_stats_init(net); 3427 } 3428 3429 static void __net_exit smc_net_stat_exit(struct net *net) 3430 { 3431 smc_stats_exit(net); 3432 } 3433 3434 static struct pernet_operations smc_net_ops = { 3435 .init = smc_net_init, 3436 .exit = smc_net_exit, 3437 .id = &smc_net_id, 3438 .size = sizeof(struct smc_net), 3439 }; 3440 3441 static struct pernet_operations smc_net_stat_ops = { 3442 .init = smc_net_stat_init, 3443 .exit = smc_net_stat_exit, 3444 }; 3445 3446 static int __init smc_init(void) 3447 { 3448 int rc; 3449 3450 rc = register_pernet_subsys(&smc_net_ops); 3451 if (rc) 3452 return rc; 3453 3454 rc = register_pernet_subsys(&smc_net_stat_ops); 3455 if (rc) 3456 goto out_pernet_subsys; 3457 3458 rc = smc_ism_init(); 3459 if (rc) 3460 goto out_pernet_subsys_stat; 3461 smc_clc_init(); 3462 3463 rc = smc_nl_init(); 3464 if (rc) 3465 goto out_ism; 3466 3467 rc = smc_pnet_init(); 3468 if (rc) 3469 goto out_nl; 3470 3471 rc = -ENOMEM; 3472 3473 smc_tcp_ls_wq = alloc_workqueue("smc_tcp_ls_wq", WQ_PERCPU, 0); 3474 if (!smc_tcp_ls_wq) 3475 goto out_pnet; 3476 3477 smc_hs_wq = alloc_workqueue("smc_hs_wq", WQ_PERCPU, 0); 3478 if (!smc_hs_wq) 3479 goto out_alloc_tcp_ls_wq; 3480 3481 smc_close_wq = alloc_workqueue("smc_close_wq", WQ_PERCPU, 0); 3482 if (!smc_close_wq) 3483 goto out_alloc_hs_wq; 3484 3485 rc = smc_core_init(); 3486 if (rc) { 3487 pr_err("%s: smc_core_init fails with %d\n", __func__, rc); 3488 goto out_alloc_wqs; 3489 } 3490 3491 rc = smc_llc_init(); 3492 if (rc) { 3493 pr_err("%s: smc_llc_init fails with %d\n", __func__, rc); 3494 goto out_core; 3495 } 3496 3497 rc = smc_cdc_init(); 3498 if (rc) { 3499 pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc); 3500 goto out_core; 3501 } 3502 3503 rc = proto_register(&smc_proto, 1); 3504 if (rc) { 3505 pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc); 3506 goto out_core; 3507 } 3508 3509 rc = proto_register(&smc_proto6, 1); 3510 if (rc) { 3511 pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc); 3512 goto out_proto; 3513 } 3514 3515 rc = sock_register(&smc_sock_family_ops); 3516 if (rc) { 3517 pr_err("%s: sock_register fails with %d\n", __func__, rc); 3518 goto out_proto6; 3519 } 3520 INIT_HLIST_HEAD(&smc_v4_hashinfo.ht); 3521 INIT_HLIST_HEAD(&smc_v6_hashinfo.ht); 3522 3523 rc = smc_ib_register_client(); 3524 if (rc) { 3525 pr_err("%s: ib_register fails with %d\n", __func__, rc); 3526 goto out_sock; 3527 } 3528 rc = smc_inet_init(); 3529 if (rc) { 3530 pr_err("%s: smc_inet_init fails with %d\n", __func__, rc); 3531 goto out_ib; 3532 } 3533 rc = bpf_smc_hs_ctrl_init(); 3534 if (rc) { 3535 pr_err("%s: bpf_smc_hs_ctrl_init fails with %d\n", __func__, 3536 rc); 3537 goto out_inet; 3538 } 3539 static_branch_enable(&tcp_have_smc); 3540 return 0; 3541 out_inet: 3542 smc_inet_exit(); 3543 out_ib: 3544 smc_ib_unregister_client(); 3545 out_sock: 3546 sock_unregister(PF_SMC); 3547 out_proto6: 3548 proto_unregister(&smc_proto6); 3549 out_proto: 3550 proto_unregister(&smc_proto); 3551 out_core: 3552 smc_core_exit(); 3553 out_alloc_wqs: 3554 destroy_workqueue(smc_close_wq); 3555 out_alloc_hs_wq: 3556 destroy_workqueue(smc_hs_wq); 3557 out_alloc_tcp_ls_wq: 3558 destroy_workqueue(smc_tcp_ls_wq); 3559 out_pnet: 3560 smc_pnet_exit(); 3561 out_nl: 3562 smc_nl_exit(); 3563 out_ism: 3564 smc_clc_exit(); 3565 smc_ism_exit(); 3566 out_pernet_subsys_stat: 3567 unregister_pernet_subsys(&smc_net_stat_ops); 3568 out_pernet_subsys: 3569 unregister_pernet_subsys(&smc_net_ops); 3570 3571 return rc; 3572 } 3573 3574 static void __exit smc_exit(void) 3575 { 3576 static_branch_disable(&tcp_have_smc); 3577 smc_inet_exit(); 3578 sock_unregister(PF_SMC); 3579 smc_core_exit(); 3580 smc_ib_unregister_client(); 3581 smc_ism_exit(); 3582 destroy_workqueue(smc_close_wq); 3583 destroy_workqueue(smc_tcp_ls_wq); 3584 destroy_workqueue(smc_hs_wq); 3585 proto_unregister(&smc_proto6); 3586 proto_unregister(&smc_proto); 3587 smc_pnet_exit(); 3588 smc_nl_exit(); 3589 smc_clc_exit(); 3590 unregister_pernet_subsys(&smc_net_stat_ops); 3591 unregister_pernet_subsys(&smc_net_ops); 3592 rcu_barrier(); 3593 } 3594 3595 module_init(smc_init); 3596 module_exit(smc_exit); 3597 3598 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>"); 3599 MODULE_DESCRIPTION("smc socket address family"); 3600 MODULE_LICENSE("GPL"); 3601 MODULE_ALIAS_NETPROTO(PF_SMC); 3602 /* 256 for IPPROTO_SMC and 1 for SOCK_STREAM */ 3603 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 256, 1); 3604 #if IS_ENABLED(CONFIG_IPV6) 3605 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET6, 256, 1); 3606 #endif /* CONFIG_IPV6 */ 3607 MODULE_ALIAS_GENL_FAMILY(SMC_GENL_FAMILY_NAME); 3608